Lodrik Adam, Sofia Benczédi, Stefan Favre, Delia Fuchs
Published
December 4, 2023
1 Introduction
1.1 Overview and Motivation
The global significance of the SDGs is our basis. The adoption of the SDGs by the United Nation in 2015 marked a significant global commitment to address pressing issues such as poverty, inequality, climate, change, and more. The fact that these goals were unanimously adopted by 193 member states underscores their importance. This prompted us to ask ourselves, can we evaluate the progress? What has really been done so far? Although the SDGs have attracted considerable attention and backing, it is essential to evaluate the events preceding and following their implementation. Understanding the actions taken and progress made is essential in determining if these global commitments are resulting in tangible enhancements to individuals’ lives. By examining the evolution of all countries and their respective contributions towards achieving the SDGs, we can develop a comprehensive understanding of collective efforts and identify potential disparities or gaps.
1.2 Related Work
1.3 Research questions
Focus on factors: What can explain the state of the countries regarding sustainable development? (we will analyse different factors: scores from the human freedom index, GDP per capita, military expenditures in % of GDP/government expenditure, unemployment rate, internet usage). See data description for more precise information about the factors.
Focus on time: How has the adoption of the SDGs in 2015 influenced the achievement of SDGs? (we want to compare the achievement (SDG scores: there are scores calculated even before the adoption) of the different countries before and after 2015 to see if the adoption of SDG gave a real “push” to sustainable development)
Focus on events: Is the evolution in sustainable development influenced by uncontrollable events, such as economic crisis, health crises and natural disasters? (we will analyse the impact of the COVID, natural disasters and conflicts (# deaths, damages, etc.) on the SDG scores). See data description for more precise information about how the impact of these events are materialized into data.
Focus on relationship between SDGs: How are the different SDGs linked? (We want to see if some SDGs are linked in the fact that a high score on one implies a high score on the other, and thus if we can make groups of SDGs that are comparable in that way).
2 Data
2.1 Sources
We are collecting our Data from the sustainability development report (SDG), the international labour organization (ILOSTAT), the World Bank, Our world in data, the CATO institute, one from Kaggle (disasters: we couldn’t find relevant accessible information from somewhere else) and GitHub. We found different datasets containing useful information in relation with the SDGs. The details about these data and the links are presented in the next section. Utilizing the kableExtra package, we provide a comprehensive list and corresponding links to our sources, as outlined below:
During the wrangling process, we added data to our table (D1_1_SDG) from different other datasets and match them based on the country code, and the year. The tables below show all the variables present in our 9 databases. We will then merge them to have our final table for the analysis.
D1_1_SDG
Our primary database focuses on the Sustainable Development Goals (SDG). Below is a table summarizing the key variables included:
Variable Name
Explanation
code
Country code (ISO)
country
Country name
year
Year of the observation (2000-2022)
overallscore
Overall score on all 17 SDGs (the score are % of achievement of the goals determined by the UN based on several indicators)
goal1:goal17
Score on each SDG except SDG 14 (16 variables)
population
Population of the country
D2_2_Unemployment_rate
Variable Name
Explanation
code
Country code (ISO)
country
Country name
year
Year of the observation (2000-2022)
unemployment.rate
Unemployment rate (% of the population 15 years old and older)
D3_0_GDP_per_capita
Variable Name
Explanation
code
Country code (ISO)
country
Country name
year
Year of the observation (2000-2022)
GDPpercapita
GDP per capita
D3_1_Military_expenditure_percent_GDP
Variable Name
Explanation
code
Country code (ISO)
country
Country name
year
Year of the observation (2000-2022)
MilitaryExpenditurePercentGDP
Military expenditures in percentage of GDP
D4_0_Internet_usage
Variable Name
Explanation
code
Country code (ISO)
country
Country name
year
Year of the observation (2000-2022)
internet.usage
Internet usage (% of the population)
D5_0_Human_freedom_index
Variable Name
Explanation
code
Country code (ISO)
country
Country name
year
Year of the observation (2000-2022)
region
Part of the world, group of countries (e.g. Eastern Europe, Dub-Saharan Africa, South Asia, etc.)
pf_law
Rule of law, mean score of: Procedural justice, Civil, justice, Criminal justice, Rule of law (V-Dem)
pf_security
Security and safety, mean score of: Homicide, Disappearances conflicts, terrorism
pf_movement
Freedom of movement (V-Dem), Freedom of movement (CLD)
pf_religion
Freedom of religion, Religious organization, repression
pf_assembly
Civil society entry and exit, Freedom of assembly, Freedom to form/run political parties, Civil society repression
pf_expression
Direct attacks on the press, Media and expression (V-Dem), Media and expression (Freedom House), Media and expression (BTI), Media and expression (CLD)
Government consumption, Transfers and subsidies, Government investment, Top marginal tax rate, State ownership of assets
ef_legal
Judicial independence, Impartial courts, Protection of property rights, Military interference Integrity of the legal system Legal enforcementof contracts, Regulatory costs, Reliability of police
ef_money
Money growth, Standard deviation of inflation, Inflation: Most recent year, Freedom to own foreign currency
ef_trade
Tariffs, Regulatory trade barriers, Black-market exchange rates, Movement of capital and people
ef_regulation
Credit market regulations, Labor market regulations, Business regulations
D6_0_Disaters
Variable Name
Explanation
code
Country code (ISO)
country
Country name
year
Year of the observation (2000-2022)
continent
Continents touched by the disasters such as floods, ouragan
total_deaths
Number of deaths caused by disasters
no_injured
Number of injured caused by disasters
no_affected
Number of affected caused by disasters
no_homeless
Number of homeless caused by disasters
total_affected
Total number of affected caused by disasters
total_damages
Total of infrastructure damages
D7_0_COVID
Variable Name
Explanation
code
Country code (ISO)
country
Country name
year
Year of the observation (2000-2022)
deaths_per_million
Number of people dead due to COVID
cases_per_million
Number of COVID cases
stringency
Government Response Stringency Index: composite measure based on 9 response indicators including school closures, workplace closures, and trave
D8_0_Conflicts
Variable Name
Explanation
code
Country code (ISO)
country
Country name
year
Year of the observation (2000-2022)
ongoing
Variable coded 1 for more than 25 deaths in intrastate conflict and 0 otherwise according to UCDP/PRIO Armed Conflict Dataset 17.1.
sum_deaths
Best estimate of deaths in all categories of violence (non-state, one-sided and state-based) recorded by the Uppsala Conflict Data Program in the country based on the UCDP GED dataset (unpublished 2016 data). The location of these events is used for estimating the extent of violence.
pop_affected
Share of population affected by violence in percentage (0 to 100) measured as described above based on population data from CIESIN, the PRIO-GRID structure as well as UCDP GED.
area_affected
Area affected by conflict
maxintensity
Two different intensity levels are coded: minor armed conflicts (1) and wars (2), Takes the max intensity of conflict in the country so that it is coded 2 if there is at least one war (>=1000 deaths in intrastate conflict) during the year. Data from UCDP/PRIO Armed Conflict Dataset 17.1.
2.3 Wrangling/cleaning
To accommodate the large scale of the datasets, we pre-cleaned each one prior to merging. This streamlined the process, simplifying the cleaning of the final, combined dataset.
2.3.1 Dataset on SDG
This is our main dataset, that we clean in order to keep the columns containing the following information: country name, country code, year, population, overall score and the SDGs scores.
We start by importing the data and converting it into a DataFrame. Next, we rename the columns and convert the scores into numeric variables.
propmissing <-numeric(length(D1_0_SDG))for (i in1:length(D1_0_SDG)){ proportion <-mean(is.na(D1_0_SDG[[i]])) propmissing[i] <- proportion}variable_names <-colnames(D1_0_SDG)prop_missing_data <-data.frame(variable = variable_names, prop_missing = propmissing)ggplot(prop_missing_data, aes(x = variable, y = prop_missing)) +geom_bar(stat ="identity", fill ="skyblue", color ="black") +labs(title ="NAs in the main dataset",x ="Variable",y ="Proportion of Missing Values") +theme_minimal()+coord_flip()
Observing that the ‘population’ column contains numerous NAs, we investigate and discover that missing values are common, as some observations represent regions, not countries. Therefore, we can safely exclude these observations.
Code
SDG0 <- D1_0_SDG |>group_by(code) |>select(population) |>summarize(NaPop =mean(is.na(population))) |>filter(NaPop !=0)ggplot(SDG0, aes(x = code, y = NaPop)) +geom_bar(stat ="identity", fill ="lightgreen", color ="black") +labs(title ="NAs in population variable are for regions and not countries",x ="Code",y ="Proportion of Missing Values") +theme_minimal() +theme(axis.text.x =element_text(angle =45, hjust =1))D1_0_SDG <- D1_0_SDG %>%filter(!str_detect(code, "^_"))
Now, there are no missing values in the ‘population’ variable, and we observe that it contains information on 166 countries.
We notice that NAs are present in only three SDG scores: 1, 10, and 14. Additionally, when a country has NAs, they occur across all years or not at all. Consequently, we decide to conduct further investigations on these three SDG scores to determine whether to include them in our analysis.
For goal 1, there are only 9.04% missing values in 15 different countries. Goal 1 being “End poverty”, we decide to keep it and only remove the countries with no information for the analysis.
For goal 10, there are only 10.2% missing values in 17 different countries. Goal 10 being “reduced inequalities”, we decide to keep it and only remove the countries with no information for the analysis.
For goal 14, there are 24.1% missing values in 40 different countries. Goal 14 being “life under water”, we decide not to keep it, because other SDG such as “life on earth” and “clean water” already treat similar subjects.
We will work with various datasets and merge them using the country code and year as key identifiers. To ensure accurate matching, we first verify that country names are encoded in UTF-8 format. Then, we standardize the names of the countries (requiring a custom match for Turkey) and the country codes, utilizing the countrycode library. Additionally, we compile a list of all country codes from the main database to filter the other datasets. Lastly, we complete the database to include all possible “country, year” combinations, ensuring the total number of rows remains unchanged.
For this first dataset, we reduced the size from 4,140 observations across 120 variables to 3,818 observations for 21 variables.
As said, this is now our main dataset. All subsequent datasets will be merged with this dataset. Therefore, for all the following datasets, we will make sure that we only keep data for the same countries and years as in this dataset. We have a total of 166 countries and the years range from 2000 to 2022.
2.3.2 Dataset on Unemployment rate
In this dataset, the initial step involves importing the data. Next, we ensure that the names and codes of the countries are formatted in UTF-8, preventing any discrepancies due to mismatches in country names. Following this, we modify the column names and filter the data to include only the relevant countries and years, specifically the years 2000 to 2022, covering 166 countries from our primary dataset.
For this first dataset, we reduced the size from 82,800 observations across 8 variables to 3,812 observations for 5 variables.
2.3.3 Dataset on GDP military Expenditures
We have three different databases which contain information on each countries over the years. Each year represent one variable. We want to extract three variables for our analysis: GDP per capita, military expenditures in percentage of the GDP and military expenditures in percentage of government expenditures.
After importing the data, we fill in the missing country codes using the column Indicator.Name, because we realized after some manipulations, that some of the country codes were false, but the next column contained the right ones.
Code
fill_code <-function(data){ data <- data %>%mutate(Country.Code =ifelse(!grepl("^[A-Z]{3}$", Country.Code), Indicator.Name, Country.Code))}
We create a set of functions that we will apply to each database. First, remove the variables that we don’t need, which are the years before 2000. Second, make sure that the values are numeric and rename the year variables (because they all had an “X” before year number). Third, transform the database from wide to long, in order to match the main database. Fourth, transform the year variable into an integer variable and rearrange and rename the columns to match the ones of the other databases. Then, we apply these transformations to the three databases.
Code
remove <-function(data){ years <-seq(1960, 1999) removeyears <-paste("X", years, sep ="") data <- data[, !(names(data) %in%c("Indicator.Name", "Indicator.Code", "X", removeyears))]}makenum <-function(data) {for (i in2000:2022) { year <-paste("X", i, sep ="") data[[year]] <-as.numeric(data[[year]]) }return(data)}renameyear <-function(data) {for (i in2000:2022) { varname <-paste("X", i, sep ="")names(data)[names(data) == varname] <-gsub("X", "", varname) }return(data)}wide2long <-function(data) { data <-pivot_longer(data, cols =-c("Country.Name", "Country.Code"), names_to ="year", values_to ="data")return(data)}yearint <-function(data) { data$year <-as.integer(data$year)return(data)}nameorder <-function(data) {colnames(data) <-c("country", "code", "year", "data") data <- data %>%select(c("code", "country", "year", "data"))}cleanwide2long <-function(data){ data <-fill_code(data) data <-remove(data) data <-makenum(data) data <-renameyear(data) data <-wide2long(data) data <-yearint(data) data <-nameorder(data)}GDPpercapita <-cleanwide2long(GDPpercapita)MilitaryExpenditurePercentGDP <-cleanwide2long(MilitaryExpenditurePercentGDP)MiliratyExpenditurePercentGovExp <-cleanwide2long(MiliratyExpenditurePercentGovExp)
We rename the colums with the main information, standardize the country code and remove the countries that are not in our main database. We see that all the 166 countries are there.
There were only 157 countries that were both in the main SDG dataset and in these 3 datasets, but we suspected that some of the missing countries were in the database but not rightly matched. Indeed, Bahamas was in the database but instead of the code “BHS” there was “The”, for “COD” it was “Dem. Rep.”, for “COG” it was “Rep”, etc. We remarked that the code is in another column of the initial database: “Indicator.Name”. We went back to the initial database and before cleaning it we put the right codes (as seen above) and after rerunning the code we see that we have all our 166 countries from the initial dataset.
We run a first round of investigation of the missing values and find that we have 16.4% for MiliratyExpenditurePercentGovExp, 12.9% for MilitaryExpenditurePercentGDP and 1.31% for GDPpercapita.
For GDPpercapita, only two countries (SOM and SSD) have a lot of missing values and in total 11 countries countries have missing values.
Code
GDPpercapita1 <- GDPpercapita %>%group_by(code) %>%summarize(NaGDP =mean(is.na(GDPpercapita))) %>%filter(NaGDP !=0)ggplot(GDPpercapita1, aes(x =reorder(code, NaGDP), y = NaGDP, fill = code)) +geom_bar(stat ="identity", fill="#FFEDCC", color="black") +labs(title ="11 countries with NAs for GDP per capita",x ="Code",y ="Proportion of Missing Values") +theme_minimal() +theme(axis.text.x =element_text(angle =45, hjust =1))
We plot the evolution of GDPpercapita avec the years for each country containing missing values and distinguish the percentage of missing values with colors.
Code
filtered_data_GDP <- GDPpercapita %>%filter(code %in% GDPpercapita1$code) # countries with NAsfiltered_data_GDP <- filtered_data_GDP %>%group_by(code) %>%mutate(PercentageMissing =mean(is.na(GDPpercapita))) %>%# column % NAsungroup()Evol_Missing_GDP <-ggplot(data = filtered_data_GDP) +geom_point(aes(x = year, y = GDPpercapita, color =cut(PercentageMissing,breaks =c(0, 0.1, 0.2, 1),labels =c("0-10%", "10-20%", "30-100%")))) +labs(title ="Evolution of GDP per capita over time", x ="Year", y ="GDP per capita") +scale_color_manual(values =c("0-10%"="blue", "10-20%"="green", "30-100%"="black"),labels =c("0-10%", "10-20%", "30-100%")) +guides(color =guide_legend(title ="% missings")) +facet_wrap(~ code, nrow =4)print(Evol_Missing_GDP)
For the countries with less than 30% of missing values and a linear evolution in time, we fill the missing values using linear interpolation.
Code
list_code <-c("AFG", "BTN", "CUB", "STP", "TKM")for (i in list_code) { country_data <- GDPpercapita %>%filter(code == i) interpolated_data <-na.interp(country_data$GDPpercapita) GDPpercapita[GDPpercapita$code == i, "GDPpercapita"] <- interpolated_data}
2.3.3.2 Military expenditures in percentage of GDP
For MilitaryExpenditurePercentGDP, 12 countries have 100% of missing values. We further investigate and keep them for now, knowing that some of these coutries may also have many missing values in the other databases when wee merge everything and will be dropped later.
Code
MilitaryExpenditurePercentGDP1 <- MilitaryExpenditurePercentGDP %>%group_by(code) %>%summarize(NaMil1 =round(mean(is.na(MilitaryExpenditurePercentGDP)),3)) %>%filter(NaMil1 !=0)ggplot(MilitaryExpenditurePercentGDP1, aes(x =reorder(code, NaMil1), y = NaMil1, fill = code)) +geom_bar(stat ="identity", fill="#FFC0CB", color="black") +labs(title ="Military expenditures in % of GDP: lot of countries have NAs",x ="Code",y ="Proportion of Missing Values") +theme_minimal() +theme(axis.text.x =element_text(angle =90, hjust =1))
We plot the evolution of MilitaryExpenditurePercentGDP along the years for each country containing missing values and distinguish the percentage of missing values with colors.
Code
filtered_data_Mil1 <- MilitaryExpenditurePercentGDP %>%filter(code %in% MilitaryExpenditurePercentGDP1$code) # countries with NAsfiltered_data_Mil1 <- filtered_data_Mil1 %>%group_by(code) %>%mutate(PercentageMissing =mean(is.na(MilitaryExpenditurePercentGDP))) %>%# Column % NAsungroup()Evol_Missing_Mil1 <-ggplot(data = filtered_data_Mil1) +geom_line(aes(x = year, y = MilitaryExpenditurePercentGDP, color =cut(PercentageMissing,breaks =c(0, 0.1, 0.2, 0.3, 1),labels =c("0-10%", "10-20%", "20-30%", "30-100%")))) +labs(title ="Military expenditure in % of GDP over time", x ="Years from 2000 to 2022", y ="GDP per capita") +scale_color_manual(values =c("0-10%"="blue", "10-20%"="green", "20-30%"="red", "30-100%"="black"),labels =c("0-10%", "10-20%", "20-30%", "50-100%")) +guides(color =guide_legend(title ="% missings")) +facet_wrap(~ code, nrow =6) +theme(strip.text =element_text(size =6)) +scale_x_continuous(breaks =NULL) +scale_y_continuous(breaks =NULL)print(Evol_Missing_Mil1)
For the countries with less than 30% of missing values and a linear evolution in time, we fill the missing values using linear interpolation.
2.3.3.3 Military expenditures in percentage of governement expenditures
For MilitaryExpenditurePercentGovExp, 17 countries have 100% of missing values. We further investigate and keep them for now, knowing that some of these coutries may also have many missing values in the other databases when wee merge everything and will be dropped later.
Code
MiliratyExpenditurePercentGovExp1 <- MiliratyExpenditurePercentGovExp %>%group_by(code) %>%summarize(NaMil2 =round(mean(is.na(MiliratyExpenditurePercentGovExp)),3)) %>%filter(NaMil2 !=0)ggplot(MiliratyExpenditurePercentGovExp1, aes(x =reorder(code, NaMil2), y = NaMil2, fill = code)) +geom_bar(stat ="identity", fill="#E6E6FA", color="black") +labs(title ="Military expenditures in % of government expenditures: lot of countries have NAs",x ="Code",y ="Proportion of Missing Values") +theme_minimal() +theme(axis.text.x =element_text(angle =90, hjust =1, size=8))
We plot the evolution of MilitaryExpenditurePercentGovExp along the years for each country containing missing values and distinguish the percentage of missing values with colors.
Code
filtered_data_Mil2 <- MiliratyExpenditurePercentGovExp %>%filter(code %in% MiliratyExpenditurePercentGovExp1$code) # Countries with NAsfiltered_data_Mil2 <- filtered_data_Mil2 %>%group_by(code) %>%mutate(PercentageMissing =mean(is.na(MiliratyExpenditurePercentGovExp))) %>%# Column % NAsungroup()Evol_Missing_Mil2 <-ggplot(data = filtered_data_Mil2) +geom_line(aes(x = year, y = MiliratyExpenditurePercentGovExp, color =cut(PercentageMissing,breaks =c(0, 0.1, 0.2, 0.3, 1),labels =c("0-10%", "10-20%", "20-30%", "30-100%")))) +labs(title ="Military expenditure in % of government expenditures over time", x ="Year from 2000 to 2022", y ="GDP per capita") +scale_color_manual(values =c("0-10%"="blue", "10-20%"="green", "20-30%"="red", "30-100%"="black"),labels =c("0-10%", "10-20%", "20-30%", "50-100%")) +guides(color =guide_legend(title ="% missings")) +facet_wrap(~ code, nrow =7) +theme(strip.text =element_text(size =6)) +scale_x_continuous(breaks =NULL) +scale_y_continuous(breaks =NULL)print(Evol_Missing_Mil2)
For the countries with less than 30% of missing values and a linear evolution in time, we fill the missing values using linear interpolation.
We now look again at the percentage of missing values for the trhee databases: 14.49% for MiliratyExpenditurePercentGovExp, 11.6% for MilitaryExpenditurePercentGDP and 1.07% for GDPpercapita
To prepare the dataset on internet usage in the world to be merge with the other data, we first, import the data. Then, we keep only the year that we are interested in (2000 to 2022). We also rename the column and keep only the country that match the list of the countries in the main dataset on the SDG.
For this first dataset, we reduced the size from 6,570 observations across 4 variables to 3,433 observations for 4 variables.
2.3.5 Dataset on human freedom index
After importing the data from the CATO Institute website, we noticed that even if the file was called “Human Freedom Index 2022”, the available observations were only going from 2000 up to 2020. We have decided first to modify it in order to match our other datasets, by renaming/encoding/standardizing the columns containing the country names.
Code
data <-read.csv(here("scripts", "data", "human-freedom-index-2022.csv"))#data in tibble datatibble <-tibble(data)# Rename the column countries into country to match the other datbasesnames(datatibble)[names(datatibble) =="countries"] <-"country"# Make sure the encoding of the country names are UTF-8datatibble$country <-iconv(datatibble$country, to ="UTF-8", sub ="byte")# standardize country namesdatatibble <- datatibble %>%mutate(country =countrycode(country, "country.name", "country.name"))
Once done, we could verify which countries were or were not present between these observations and our main SDG dataset. We have decided to keep the ones that were matching between the two datasets.
Code
# Merge by country namedatatibble <- datatibble %>%left_join(D1_0_SDG_country_list, by ="country")datatibble <- datatibble %>%filter(code %in% list_country)(length(unique(datatibble$code)))#> [1] 159# See which ones are missinglist_country_free <-c(unique(datatibble$code))(missing <-setdiff(list_country, list_country_free))#> [1] "AFG" "CUB" "MDV" "STP" "SSD" "TKM" "UZB"# Turkey was missing but present in the initial database (it was a problem when stadardizing the country names of D1_0SDG_country_list that we corrected) and the other missing countries are:"AFG" "CUB" "MDV" "STP" "SSD" "TKM" "UZB" D5_0_Human_freedom_index <- datatibble
Then, we noticed that there were a lot of columns that were not important for us, as we had 141 variables taken into account. So we have decided to keep the ones that refers to the countries informations (such as code, year, ..) and their human freedom scores per category (pf for personnal freedom, ef for economical freedom).
Code
# erasing useless columns to keep only the general ones. D5_0_Human_freedom_index <-select(D5_0_Human_freedom_index, year, country, region, hf_score, pf_rol, pf_ss, pf_movement, pf_religion, pf_assembly, pf_expression, pf_identity, pf_score, ef_government, ef_legal, ef_money, ef_trade, ef_regulation, ef_score, code)D5_0_Human_freedom_index <- D5_0_Human_freedom_index %>%rename(pf_law =names(D5_0_Human_freedom_index)[5], # Renames the 5th column to "pf_law"pf_security =names(D5_0_Human_freedom_index)[6] # Renames the 6th column to "pf_security" )
After renaming the columns pf_law/security for comprehension purpose, we have investigated how are distributed the NA values among the countries and the variables. After having found the percentages of missing values per country and variable, heatmaps revealed themself to be a great tool for visualizing datas.
Then, for having a better understanding of the situation, we ordered the countries having at least 1 variable containing 50% and more of missing values
Code
na_long <- na_long %>%group_by(country) %>%mutate(Count_NA_50_100 =sum(NA_Percentage >=50& NA_Percentage <=100, na.rm =TRUE)) %>%ungroup() %>%arrange(desc(Count_NA_50_100))heatmap_ordered_all <-ggplot(na_long, aes(x =reorder(country, -Count_NA_50_100), y = Variable)) +geom_tile(aes(fill = NA_Percentage), colour ="white") +scale_fill_gradient(low ="white", high ="red") +theme_minimal() +labs(title ="Heatmap of NA Percentages per Country and Variable",x ="Countries",y ="Variables",fill ="NA Percentage" ) +theme(axis.text.x =element_blank(), # Hide x-axis labelsaxis.text.y =element_text(size =9) )print(heatmap_ordered_all)
We notice that only some countries look to contain at least 50% of missing values and in addition that most of the missing values are concerning the EF variables (Economic Freedom). Now, we tried to produce another heatmap only containing the ordered countries, and also counting for each one of these country the number of variables with at least 50% of NAs.
We conclude here that 13 countries were concerned by our selection of 50% and more of missing values. When discussing between us, we came to the conclusion that among these 13 countries, a great part of them were not going to be selected because they had a lot of missing values in our main dataset too. Therefore, we have decided to merge this data with the other datasets and finish the cleaning after.
Here are the first few lines of the partialy cleaned dataset on Human Freedom Index scores:
For this first dataset, we reduced the size from 3’465 observations across 141 variables to 3339 observations for 4 variables.
2.3.6 Dataset on Disasters
For this dataset concerning the Disasters we imported the data from Kaggle as we couldn’t find the original dataset that is private coming from the EOSDIS SYSTEM, an interactive interface for browsing full-resolution, global, daily satellite images from NASA. Once we made sure that our file called “Disasters” was convert into a data frame, we selected some specific columns that we where interested in.
Because we knew that our file showed all the disasters in each country over the years (between 1970-2021) and that we wanted to focus on a specific period, we filtered our data to show the years between 2000 and 2022. Then we rearranged our data, changing the data types of all the columns and their names in order to match our other datasets.
Code
# Rearrange the columns, changed the type of data, renamed the columnsRearanged_Disasters <- Disasters %>%filter(Year >=2000& Year <=2022) %>%mutate(code =as.character(ISO),country =as.character(Country),year =as.integer(Year),continent =as.character(Continent),disaster.subgroup =as.character(Disaster.Subgroup),disaster.type =as.character(Disaster.Type),location =as.character(Location),total.deaths =as.numeric(Total.Deaths),no.injured =as.numeric(No.Injured),no.affected =as.numeric(No.Affected),no.homeless =as.numeric(No.Homeless),total.affected =as.numeric(Total.Affected),total.damages =as.numeric(Total.Damages...000.US..) )
We then grouped the data by “year”, “code”, “country” and “continent” and summarize the data. Here you can see that we re-selected specific columns as we saw that our first pre-selection was still too wide and some variables as the disaster.subgroup and disaster.type weren’t pertinent.We arranged the columns based on “code,” “country,” “year,” and “continent” to match the other datasets.
Finally we filtered our disasters data to keep only the countries that are present in our main dataset. We analysed the missing countries and identified three countries (BHR, BRN, MLT) that are unexpectedly missing.
Code
D6_0_Disasters <- D6_0_Disasters %>%filter(code %in% list_country)length(unique(D6_0_Disasters$code))#> [1] 163# Here we see which countries are missinglist_country_disasters <-c(unique(D6_0_Disasters$code))(missing <-c(missing,setdiff(list_country, list_country_disasters)))#> [1] "AFG" "CUB" "MDV" "STP" "SSD" "TKM" "UZB" "BHR" "BRN" "MLT"
Here are the first few lines of the cleaned dataset on Disasters:
This dataset contains information on the COVID19 pandemic between 2020 and 2022. The observation are by year, month, day. After importing the database, we transform the date in format YYYY-MM-DD in order to only keep the year.
We perform a first round of investigation of the missing values before aggregating the values by year. We begin with the variables “cases per million” and “deaths per million”: seeing that for each country, we have either only missing values, either a very low percentage of missing values (~1%), we can compute the sum over each year and ignore the missing values without altering the data. Indeed, where all the values are missing, the computation will return a NA. We then look at the “stringency” variable and we have 3 scenarios:
~20% missings: we ignore missing values when computing the mean to have an idea of stringency each year (because we compute the mean stringency over the year, if some days are missing, it is not a problem, it can not evoluate that fast).
all are missing: we can ignore the missing values when computing the mean, because it will still return a missing value
almost all are missing: here the mean doesn’t make sense -> we will replace the values by NAs to be coherent. The countries with this issues are: ERI, GUM, PRI and VIR. We verify if they are in our main dataset and since none of these countries are, we can ignore the issue, the lines will be remove later anyway.
We aggregate the observations of all days of a year in one observation per country using the mean.
Code
COVID1 <- COVID %>%group_by(iso_code) %>%summarize(NaDeaths =round(mean(is.na(new_deaths_per_million)),3),NaCases =round(mean(is.na(new_cases_per_million)), 3),NaStringency =round(mean(is.na(stringency_index)), 3)) %>%pivot_longer(cols =starts_with("Na"), names_to ="Variable", values_to ="NaValue")%>%filter(NaValue!=0)issue_list <-c("ERI", "GUM", "PRI", "VIR")is.element(issue_list, list_country)#> [1] FALSE FALSE FALSE FALSECOVID <- COVID %>%group_by(location, date) %>%mutate(cases_per_million =sum(new_cases_per_million, na.rm =TRUE),deaths_per_million =sum(new_deaths_per_million, na.rm =TRUE),stringency =mean(stringency_index, na.rm =TRUE) )%>%ungroup()#### Create a bubble plotplot_ly(COVID1, x=~Variable, y=~NaValue,type ="scatter",marker =list(color="blue", opacity=0.1, size =10))%>%layout(title ="NAs for the COVID variables",titlefont =list(size =20),xaxis =list(title ="Variables"),yaxis =list(title ="% NAs"),annotations =list(text ="The more the circile is blue the more countries are concerned",x =0.5, y =1.02, # Adjust the position of the subtitlexref ="paper", yref ="paper",showarrow =FALSE,font =list(size =16) # Adjust the size of the subtitle font ) )
Now that all the variables of interest are aggregated by year, we remove all the variables that we don’t need and rename all the remaining variables to match the main dataset.
We remove the years that exceed 2022, we make sure that the country codes are all iso codes with 3 letters (we observe that sometimes they are preceded by “OWID_”) and we standardize the country codes.
We remove the observations of countries that aren’t in our main dataset on SDGs and find that all the 166 countries that we have in the main SDG dataset are also in this one.
We perform a second round of missing values investigation and find out that there are no missing values except for the stringency, where there are 4.19%. Either all values are missing for one country, or 50% are missing, so these 7 countries won’t be included when analyzing the effect of stringency on the SDG scores.
Code
mean(is.na(COVID$cases_per_million))#> [1] 0mean(is.na(COVID$deaths_per_million))#> [1] 0mean(is.na(COVID$stringency))#> [1] 0.0419COVID4 <- COVID %>%group_by(code) %>%summarize(NaCOVID =mean(is.na(stringency))) %>%filter(NaCOVID !=0)ggplot(COVID4, aes(x =reorder(code, NaCOVID), y = NaCOVID)) +geom_bar(stat ="identity", fill ="lightgreen", color ="black") +labs(title ="Stringency has either 100% or 50% NAs or 0%",x ="Code",y ="Proportion of Missing Values") +theme_minimal() +theme(axis.text.x =element_text(angle =45, hjust =1))D7_0_COVID <- COVID
Here are the first few lines of the cleaned dataset on COVID19:
For our conflicts dataset, we imported the data from “The World Banck” data catalog. Once we made sure that our file called “Disasters” was convert into a data frame, we selected some specific columns that we where interested in.
Our file showed all the Conflicts and consequences per country over the years (between 2000-2016). We couldn’t find a better and more complete dataset, As we consider conflicts as events, we will only take into account results between 2000 and 2016. Then we rearranged our data, changing the data types of all the columns and their names in order to match our other datasets. We grouped the data by ” year”, “country”, re-selected some variables and summarize the data.
Code
Rearanged_Conflicts <- Conflicts %>%filter(year >=2000& year <=2022)%>%mutate(ongoing =as.integer(ongoing),country =as.character(country),year =as.integer(year),gwsum_bestdeaths =as.numeric(gwsum_bestdeaths),pop_affected =as.numeric(pop_affected),area_affected =as.numeric(area_affected),maxintensity =as.numeric(maxintensity), )# Group the data by "year", "country" and summarize the dataConflicts <- Rearanged_Conflicts %>%group_by(year, country) %>%summarize(ongoing =sum (ongoing, na.rm =TRUE),sum_deaths =sum(gwsum_bestdeaths, na.rm =TRUE),pop_affected =sum(pop_affected, na.rm =TRUE),area_affected =sum(area_affected, na.rm =TRUE),maxintensity =sum(maxintensity, na.rm =TRUE), )
After we Selected specific columns from the summarized data and arrange the data by our specified columns. To make our dataset compatible with the main one and let the merging face succeed, we dd some adjustment concerning the country names’ to ensure the compatibility. Then we standardize and merge by country names to finally rearrange the data to retain only the countries present in our main dataset. Note that in the end we can see that only one country is missing that wasn’t in the initial conflicts database: BLR
Code
conflicts <- Conflicts %>%select(country, year, ongoing, sum_deaths, pop_affected, area_affected, maxintensity) %>%arrange(country, year)conflicts$country <-iconv(conflicts$country, to ="UTF-8", sub ="byte")conflicts <- conflicts %>%mutate(country =countrycode(country, "country.name", "country.name"))conflicts <- conflicts %>%left_join(D1_0_SDG_country_list, by ="country")conflicts <- conflicts %>%select(code, country, year, ongoing, sum_deaths, pop_affected, area_affected, maxintensity) %>%arrange(code, country, year)D8_0_Conflicts <- conflicts %>%filter(code %in% list_country)(length(unique(conflicts$code)))#> [1] 166# See which countries are missinglist_country_conflicts <-c(unique(conflicts$code))(missing <-c(missing, setdiff(list_country, list_country_conflicts)))#> [1] "AFG" "CUB" "MDV" "STP" "SSD" "TKM" "UZB" "BHR" "BRN" "MLT"#> [11] "BLR"
Here are the first few lines of the cleaned dataset on Conflicts:
Since we took the information on the continent and region from databases that are not the main one, we complete these inforamtion for the whole final dataset.
Final structure of our merged database: each country of the 166 countries from D1_1_SDG are observed each year from 2000 to 2022, thus each row has a key composed of (code, year) that uniquely identifies an observation. The other columns are the variables listed above. Due to some countries having a lot of missing information we will have to eliminate some of them, but we will still have more than 2000 rows in our database.
2.3.11 Treatment of missing values
We load our final database and we vizualize the missing values.
Code
all_Merge <-read.csv(here("scripts","data","all_Merge.csv"))all_Merge <- all_Merge %>%select(-c(X))# Create a dataframe with the goals without NAs summarize in one column to simplify the visualizationgoal_vars <- all_Merge %>%select(starts_with("goal")) %>%filter_all(all_vars(!is.na(.))) %>%colnames()to_plot_missing <- all_Merge %>%mutate(Goals_without_NAs =rowSums(!is.na(select(., all_of(goal_vars))))) %>%select(-c(goal2, goal3, goal4, goal5, goal6, goal7, goal8, goal9, goal11, goal12, goal13, goal15, goal16, goal17))vis_dat(to_plot_missing, warn_large_data =FALSE) +scale_fill_brewer(palette ="Paired") +theme(axis.text.x =element_text(angle =90, size =6),legend.text =element_text(size =8), # Adjust the size of legend textlegend.title =element_text(size =10) )
We subset our database according to the data that we will need in order to answer the different questions. This will help us dealing with the missing values.
For question 1, we only keep the years until 2020, because most of the explanatory variables that we want to use (those coming from the human freedom index) only have values until 2020.
For question 3, we create 3 distinct databases according to the different type of event that we wwill analyse: disasters, COVID19 and conflicts. For the disasters, we only keep the years until 2021, because after this date, we don’t have data. For the conflicts, we only keep the years until 2016, because after this date, we don’t have data.
We begin by visualizing the missing values. To have a less messy graph we group all the goals wihtout NAs into one single variable.
Code
# Create a dataframe with the goals without NAs summarize in one column to simplify the visualizationvariable_names <-names(data_question1)missing_percentages <-sapply(data_question1, function(col) mean(is.na(col)) *100)missing_data_summary <-data.frame(Variable = variable_names,Missing_Percentage = missing_percentages)missing_data_summary <- missing_data_summary %>%mutate(VariableGroup =ifelse(startsWith(Variable, "goal") & Missing_Percentage ==0, "Goals without NAs", as.character(Variable)))ggplot(data = missing_data_summary, aes(x =reorder(VariableGroup, Missing_Percentage), y = Missing_Percentage, fill = Missing_Percentage)) +geom_bar(stat ="identity") +geom_text(aes(label =ifelse(Missing_Percentage >1, sprintf("%.1f%%", Missing_Percentage), ""),y = Missing_Percentage),position =position_stack(vjust =1), # Adjust vertical positioncolor ="white", # Text colorsize =2, # Text sizehjust =1.05) +labs(title ="Percentage of Missing Values by Variable",x ="Variable",y ="Missing Percentage") +theme_minimal() +theme(axis.text.y =element_text(hjust =1, size=6 ),legend.text =element_text(size =8),legend.title =element_text(size =10)) +labs(fill ="% NAs") +coord_flip()
We create a column with the number of missing values by country over all the variables, except goal 1 and goal 10 that we already discussed. We decide to remove the countries that have more than 50 missing values.
Here is the graph that allows us to visualize the countries that have missing values, how many and for which variables, when there are more than 50 NAs in total.
Code
ggplot(see_missing1_1, aes(x = num_missing , y =reorder(code, num_missing), fill = num_missing)) +geom_bar(stat ="identity") +scale_fill_gradient(low ="lightgreen", high ="darkgreen") +theme_minimal() +theme(axis.text.y =element_text(hjust =1, size=8 ),legend.text =element_text(size =8),legend.title =element_text(size =10)) +labs(title ="Number of missing values per country containing at least 50 NAs", x ="Number of Missing Values", y ="Countries")
Now, looking at the remaining countries that have missing values and there number across all variables, we decide to remove MilitaryExpenditurePercentGovExp, because it has too many missing values and it contains similar information to MilitaryExpenditurePercentGDP.We also remove hf_score, pf_score and ef_score, because there are many missing values and since these variables summarize the other ones, deleting them will not make us loose information.
Here is the ggplot that helps us to visualize the countries that have missing values after removing the countries with more than 50 NAs.
Code
ggplot(see_missing1_2, aes(x = num_missing , y =reorder(code, num_missing), fill = num_missing)) +geom_bar(stat ="identity", width =0.5) +scale_fill_gradient(low ="lightgreen", high ="darkgreen") +theme_minimal() +theme(axis.text.y =element_text(hjust =1, size=6 ),legend.text =element_text(size =8),legend.title =element_text(size =10)) +labs(title ="Number of missing values per country", x ="Number of Missing Values", y ="Countries")
We also look at patterns of missing values in the rows and see that except for the two goals with NAs that we discussed earlier, there are not much patterns. ::: {.cell layout-align=“center”}
Then, we look at the distribution of the variable per region. Seeing that all are skewed distributions, we decide to replace the missing values, where there are less than 30% missing using the median by region.
Code
question1_missing_Military <- data_question1 %>%group_by(code) %>%mutate(PercentageMissing =mean(is.na(MilitaryExpenditurePercentGDP))) %>%# Column % NAsungroup() %>%group_by(region) %>%filter(sum(PercentageMissing, na.rm =TRUE) >0)Freq_Missing_Military <-ggplot(data = question1_missing_Military) +geom_histogram(aes(x = MilitaryExpenditurePercentGDP, fill =cut(PercentageMissing,breaks =c(0, 0.1, 0.2, 0.3, 1),labels =c("0-10%", "10-20%", "20-30%", "30-100%"))),bins =30) +labs(title ="Distribution of Military expenditures in % of GDP", x ="Military expenditures in % of GDP", y ="Frequency") +scale_fill_manual(values =c("0-10%"="blue", "10-20%"="green", "20-30%"="red","30-100%"="black"), labels =c("0-10%", "10-20%", "20-30%","30-100%")) +guides(fill =guide_legend(title ="% missings")) +facet_wrap(~ region, nrow =3)print(Freq_Missing_Military)data_question1 <- data_question1 %>%group_by(code) %>%mutate(PercentageMissingByCode =mean(is.na(MilitaryExpenditurePercentGDP)) ) %>%ungroup() %>%group_by(region) %>%mutate(MedianByRegion =median(MilitaryExpenditurePercentGDP, na.rm =TRUE),MilitaryExpenditurePercentGDP =ifelse( PercentageMissingByCode <0.3&!is.na(MilitaryExpenditurePercentGDP), MilitaryExpenditurePercentGDP,ifelse(PercentageMissingByCode <0.3, MedianByRegion, MilitaryExpenditurePercentGDP) ) ) %>%select(-PercentageMissingByCode, -MedianByRegion)
We look at the evolution of the variable over time. We fill the missing values with linear interpolation, because all evolutions are in an increasing way and are almost straight lines, except for CIV that we delete.
Code
question1_missing_Internet <- data_question1 %>%group_by(code) %>%mutate(PercentageMissing =mean(is.na(internet_usage))) %>%# Column % NAsfilter(code %in% question1_missing_Internet$code)Evol_Missing_Internet <-ggplot(data = question1_missing_Internet) +geom_line(aes(x = year, y = internet_usage, color =cut(PercentageMissing,breaks =c(0, 0.1, 0.2, 0.3, 1),labels =c("0-10%", "10-20%", "20-30%", "30-100%")))) +labs(title ="Evolution of internet usage over time", x ="Years from 2000 to 2022", y ="Internet usage") +scale_color_manual(values =c("0-10%"="blue", "10-20%"="green", "20-30%"="red", "30-100%"="black"),labels =c("0-10%", "10-20%", "20-30%", "50-100%")) +guides(color =guide_legend(title ="% missings")) +scale_x_continuous(breaks=NULL)+facet_wrap(~ code, nrow =4)print(Evol_Missing_Internet)list_code <-setdiff(unique(question1_missing_Internet$code), "CIV")for (i in list_code) { country_data <- data_question1 %>%filter(code == i) interpolated_data <-na.interp(country_data$internet_usage) data_question1[data_question1$code == i, "internet_usage"] <- interpolated_data}data_question1 <- data_question1 %>%filter(code!="CIV")list_country_deleted <-c(list_country_deleted, "CIV")
2.3.11.1.4Human freedom index
2.3.11.1.4.1Personal freedom: law
The variable pf_law has (many) NAs, but only for one country: BLZ, so we decide to remove it.
Only KGZ and SRB have missing values, we plot the values over time and fill in the missing values by the year before, since there are only one and two missing values.
Code
data_question1 %>%filter(code %in%c("KGZ", "SRB")) %>%ggplot(aes(x = year, y = ef_government)) +geom_point(color ="green") +facet_wrap(~ code, nrow =1) +labs(title ="Evolution of economic freedom: government over time", x ="Years", y ="ef_gov")data_question1 <- data_question1 %>%mutate(ef_government =ifelse(code =="KGZ"& year ==2000&is.na(ef_government), ef_government[which(code =="KGZ"& year ==2001)], ef_government))data_question1 <- data_question1 %>%mutate(ef_government =ifelse(code =="SRB"& year ==2000&is.na(ef_government), ef_government[which(code =="SRB"& year ==2002)], ef_government))data_question1 <- data_question1 %>%mutate(ef_government =ifelse(code =="SRB"& year ==2001&is.na(ef_government), ef_government[which(code =="SRB"& year ==2002)], ef_government))
2.3.11.1.4.3Economic freedom: money
18 countries have missing values, but the percentage of missing values is always below 25%.
We look at the evolution of the variable over time. For the countries where this evolution is linear, we fill in the missing values using linear interpolation.
Code
question1_missing_ef_money <- data_question1 %>%group_by(code) %>%mutate(PercentageMissing =mean(is.na(ef_money))) %>%# Column % NAsfilter(code %in% question1_missing_ef_money$code)Evol_Missing_ef_money <-ggplot(data = question1_missing_ef_money) +geom_line(aes(x = year, y = ef_money, color =cut(PercentageMissing,breaks =c(0, 0.1, 0.2, 0.3, 1),labels =c("0-10%", "10-20%", "20-30%", "30-100%")))) +labs(title ="Evolution of economic freedom: money over time", x ="Years from 2000 to 2022", y ="ef_money") +scale_color_manual(values =c("0-10%"="blue", "10-20%"="green", "20-30%"="red", "30-100%"="black"),labels =c("0-10%", "10-20%", "20-30%", "50-100%")) +guides(color =guide_legend(title ="% missings")) +facet_wrap(~ code, nrow =4) +scale_x_continuous(breaks =NULL)print(Evol_Missing_ef_money)list_code <-c("ARM", "BFA", "BIH", "GEO", "KAZ", "LSO", "MDA", "MKD")for (i in list_code) { country_data <- data_question1 %>%filter(code == i) interpolated_data <-na.interp(country_data$ef_money) data_question1[data_question1$code == i, "ef_money"] <- interpolated_data}
Then, we look at the distribution of the variable per region. Seeing that all are skewed distributions, we decide to replace the missing values using the median by region.
We look at the evolution of the variable over time. For the countries where this evolution is linear, we fill in the missing values using linear interpolation.
Code
Evol_Missing_ef_trade <-ggplot(data = question1_missing_ef_trade) +geom_line(aes(x = year, y = ef_trade, color =cut(PercentageMissing,breaks =c(0, 0.1, 0.2, 0.3, 1),labels =c("0-10%", "10-20%", "20-30%", "30-100%")))) +labs(title ="Evolution of economic freedom: trade over time", x ="Years from 2000 to 2022", y ="ef_trade") +scale_color_manual(values =c("0-10%"="blue", "10-20%"="green", "20-30%"="red", "30-100%"="black"),labels =c("0-10%", "10-20%", "20-30%", "50-100%")) +guides(color =guide_legend(title ="% missings")) +facet_wrap(~ code, nrow =4) +scale_x_continuous(breaks =NULL)print(Evol_Missing_ef_trade)# Linear interpolation for "AZE", "BFA", "ETH", "GEO", "VNH"list_code <-c("AZE", "BFA", "ETH", "GEO", "VNH")for (i in list_code) { country_data <- data_question1 %>%filter(code == i) interpolated_data <-na.interp(country_data$ef_trade) data_question1[data_question1$code == i, "ef_trade"] <- interpolated_data}
Then, we look at the distribution of the variable per region. Seeing that all are skewed distributions, we decide to replace the missing values using the median by region.
We look at the evolution of the variable over time. For the countries where this evolution is linear, we fill in the missing values using linear interpolation.
Code
question1_missing_ef_regulation <- data_question1 %>%group_by(code) %>%mutate(PercentageMissing =mean(is.na(ef_regulation))) %>%filter(code %in% question1_missing_ef_regulation$code)Evol_Missing_ef_regulation <-ggplot(data = question1_missing_ef_regulation) +geom_line(aes(x = year, y = ef_regulation, color =cut(PercentageMissing,breaks =c(0, 0.1, 0.2, 0.3, 1),labels =c("0-10%", "10-20%", "20-30%", "30-100%")))) +labs(title ="Evolution of economic freedom: regulation over time", x ="Years from 2000 to 2022", y ="ef_regulation") +scale_color_manual(values =c("0-10%"="blue", "10-20%"="green", "20-30%"="red", "30-100%"="black"),labels =c("0-10%", "10-20%", "20-30%", "50-100%")) +guides(color =guide_legend(title ="% missings")) +scale_x_continuous(breaks =NULL)+facet_wrap(~ code, nrow =2)print(Evol_Missing_ef_regulation)list_code <-c("ETH", "KAZ", "MDA", "SRB")for (i in list_code) { country_data <- data_question1 %>%filter(code == i) interpolated_data <-na.interp(country_data$ef_regulation) data_question1[data_question1$code == i, "ef_regulation"] <- interpolated_data}
Then, we look at the distribution of the variable per region. Seeing that all are skewed distributions, we decide to replace the missing values using the median by region.
Now, we notice that there were only missing values for goals 1 and 10. As we did before, we have started to investigate where are located the NAs in our dataset for first goal1, then goal 10.
Code
na_count <-sapply(data_question1, function(x) sum(is.na(x)))na_count_df <-data.frame(variable =names(na_count), num_missing = na_count)na_count_df_filtered <-subset(na_count_df, num_missing >0)ggplot(na_count_df_filtered, aes(x= num_missing,y=variable, fill = num_missing)) +geom_bar(aes(fill = num_missing), stat ="identity", width =0.8, fill ='lightblue') +geom_text(aes(label = num_missing), vjust =0.5,hjust =1.1, position =position_dodge(width =0.9)) +theme_minimal() +theme(axis.text.y =element_text(hjust =1, size=10 ), legend.text =element_text(size =8),legend.title =element_text(size =10)) +labs(title ="Number of remaining missing values per variable ",x ="Number of NAs",y ="Variables")# goal1question1_missing_goal1 <- data_question1 %>%group_by(code) %>%summarize(Na_goal1 =mean(is.na(goal1)))%>%filter(Na_goal1 !=0)data_question1 <- data_question1 %>%filter(!code %in% question1_missing_goal1$code)# Update List of countries deletedlist_country_deleted <-c(list_country_deleted, "KWT","NZL","OMN","SGP","UKR")# still 42 NA values goal10
We had found that the missing values were located in only 5 countries. So we have decided to get rid of them. At this stage, there were only 42 remaining missing values. Then, same step for goal 10.
Code
#goal10question1_missing_goal10 <- data_question1 %>%group_by(code) %>%summarize(Na_goal10 =mean(is.na(goal10)))%>%filter(Na_goal10 !=0)data_question1 <- data_question1 %>%filter(!code %in% question1_missing_goal10$code)# Update List of countries deletedlist_country_deleted <-c(list_country_deleted, "GUY","TTO")
We have found the 2 lasts contries containing missing values. Now, our dataset is completely clean and ready to be used for our question 1.
2.3.11.2 Data for question 2 and 4
We create a column with the number of missing values by country over all the variables, except goal 1 and goal 10 that we already discussed. Since there are no other missing values, we stop here.
We create a column with the number of missing values by country over all the variables, except goal 1 and goal 10 that we already discussed. Since there are no other missing values, we stop here.
We create a column with the number of missing values by country over all the variables, except goal 1 and goal 10 that we already discussed. We find out that there are many missing values and here are the first few lines identifying them by country.
In this particular case, even if there are many missing values in our disaster dataset, we made the hypothesis that disaster events can not happen every year for every country given that these are uncontrollable and non-recurring events. Therefore the NAs that we encounter will become zeroes, implying that there have been no climatic disasters.
Code
data_question3_1[is.na(data_question3_1)] <-0
COVID19
We create a column with the number of missing values by country over all the variables, except goal 1 and goal 10 that we already discussed. Since there are no other missing values, we stop here.
We create a column with the number of missing values by country over all the variables, except goal 1 and goal 10 that we already discussed.Two countries have missing values, we remove them (MNE and SRB).
We display the distribution of the different SDG achievement scores, using boxplots to have an overview of the median, the range with most of the observations and the outliers.
<<<<<<< Updated upstream
=======
>>>>>>> Stashed changes
Code
data_question1 <-read.csv(here("scripts","data","data_question1.csv"))data_question24 <-read.csv(here("scripts", "data", "data_question24.csv"))data_question2 <-read.csv(here("scripts", "data", "data_question24.csv"))data_question3_1 <-read.csv(here("scripts", "data", "data_question3_1.csv"))data_question3_2 <-read.csv(here("scripts", "data", "data_question3_2.csv"))data_question3_3 <-read.csv(here("scripts", "data", "data_question3_3.csv"))Q3.1<-read.csv(here("scripts", "data", "data_question3_1.csv"))Q3.2<-read.csv(here("scripts", "data", "data_question3_2.csv"))Q3.3<-read.csv(here("scripts", "data", "data_question3_3.csv"))data <-read.csv(here("scripts", "data", "all_Merge.csv"))Correlation_overall <- data_question1 %>%select(population:ef_regulation)#### boxplots #####for goals#dev.off()# boxplot(Correlation_overall[2:18], # las = 2, # Makes the axis labels perpendicular to the axis# par(mar = c(5, 4, 4, 2) + 0.1), # Adjusts the margins to fit all labels# cex.axis = 0.7, # Reduces the size of the axis labels# cex.lab = 1, # Reduces the size of the x and y labels# notch = TRUE, # Specifies whether to add notches or not# main = "Merged goals boxplot", # Title of the boxplot# xlab = "Goals", # X-axis label# ylab = "Score") # Y-axis label#boxplot per continentdata_Q1_Africa <- data_question1 %>%filter(data_question1$continent =='Africa')data_Q1_Europe <- data_question1 %>%filter(data_question1$continent =='Europe')data_Q1_Asia <- data_question1 %>%filter(data_question1$continent =='Asia')data_Q1_Americas <- data_question1 %>%filter(data_question1$continent =='Americas')data_Q1_Oceania <- data_question1 %>%filter(data_question1$continent =='Oceania')#Africadata_Q1_Africa_long <-melt(data_Q1_Africa[,8:24])medians_AF <- data_Q1_Africa_long %>%group_by(variable) %>%summarize(median_value =median(value))medians_AF$color <-ifelse(medians_AF$median_value >75, "lightblue", ifelse(medians_AF$median_value <25, "red", 'orange'))data_Q1_Africa_long <- data_Q1_Africa_long %>%left_join(medians_AF, by ="variable")bandwidth_nrd_AF <-bw.nrd(data_Q1_Africa_long$value)AF <-ggplot(data_Q1_Africa_long, aes(x = variable, y = value, fill = color)) +geom_violin(trim =FALSE, bw = bandwidth_nrd_AF) +
<<<<<<< Updated upstream
scale_fill_identity() +labs(title ="Africa SDG goals boxplot", x ="Goals", y ="Score") +geom_boxplot(width =0.1, outlier.size =1, fill ='white') +scale_y_continuous(labels = scales::label_number()) +theme_classic() +theme(axis.text.x =element_text(angle =90, vjust =0.5, hjust =1))#Europedata_Q1_Europe_long <-melt(data_Q1_Europe[,8:24])medians_EU <- data_Q1_Europe_long %>%group_by(variable) %>%summarize(median_value =median(value))medians_EU$color <-ifelse(medians_EU$median_value >75, "lightblue", ifelse(medians_EU$median_value <25, "red", 'orange'))data_Q1_Europe_long <- data_Q1_Europe_long %>%left_join(medians_EU, by ="variable")bandwidth_nrd_EU <-bw.nrd(data_Q1_Europe_long$value)EU <-ggplot(data_Q1_Europe_long, aes(x = variable, y = value, fill = color)) +geom_violin(trim =FALSE, bw = bandwidth_nrd_EU) +scale_fill_identity() +labs(title ="European SDG goals boxplot", x ="Goals", y ="Score") +geom_boxplot(width =0.1, outlier.size =1, fill ='white') +scale_y_continuous(labels = scales::label_number()) +theme_classic() +theme(axis.text.x =element_text(angle =90, vjust =0.5, hjust =1))#Asiadata_Q1_Asia_long <-melt(data_Q1_Asia[,8:24])medians_AS <- data_Q1_Asia_long %>%group_by(variable) %>%summarize(median_value =median(value))medians_AS$color <-ifelse(medians_AS$median_value >75, "lightblue", ifelse(medians_AS$median_value <25, "red", 'orange'))data_Q1_Asia_long <- data_Q1_Asia_long %>%left_join(medians_AS, by ="variable")bandwidth_nrd_AS <-bw.nrd(data_Q1_Asia_long$value)AS <-ggplot(data_Q1_Asia_long, aes(x = variable, y = value, fill = color)) +geom_violin(trim =FALSE, bw = bandwidth_nrd_AS) +scale_fill_identity() +labs(title ="Asian SDG goals boxplot", x ="Goals", y ="Score") +geom_boxplot(width =0.1, outlier.size =1, fill ='white') +scale_y_continuous(labels = scales::label_number()) +theme_classic() +theme(axis.text.x =element_text(angle =90, vjust =0.5, hjust =1))#Americasdata_Q1_Americas_long <-melt(data_Q1_Americas[,8:24])medians_AM <- data_Q1_Americas_long %>%group_by(variable) %>%summarize(median_value =median(value))medians_AM$color <-ifelse(medians_AM$median_value >75, "lightblue", ifelse(medians_AM$median_value <25, "red", 'orange'))data_Q1_Americas_long <- data_Q1_Americas_long %>%left_join(medians_AM, by ="variable")bandwidth_nrd_AM <-bw.nrd(data_Q1_Americas_long$value)AM <-ggplot(data_Q1_Americas_long, aes(x = variable, y = value, fill = color)) +geom_violin(trim =FALSE, bw = bandwidth_nrd_AM) +scale_fill_identity() +labs(title ="American SDG goals boxplot", x ="Goals", y ="Score") +geom_boxplot(width =0.1, outlier.size =1, fill ='white') +scale_y_continuous(labels = scales::label_number()) +theme_classic() +theme(axis.text.x =element_text(angle =90, vjust =0.5, hjust =1))#Oceaniadata_Q1_Oceania_long <-melt(data_Q1_Oceania[,8:24])medians_OC <- data_Q1_Oceania_long %>%group_by(variable) %>%summarize(median_value =median(value))medians_OC$color <-ifelse(medians_OC$median_value >75, "lightblue", ifelse(medians_OC$median_value <25, "red", 'orange'))data_Q1_Oceania_long <- data_Q1_Oceania_long %>%left_join(medians_OC, by ="variable")bandwidth_nrd_OC <-bw.nrd(data_Q1_Oceania_long$value)OC <-ggplot(data_Q1_Oceania_long, aes(x = variable, y = value, fill = color)) +geom_violin(trim =FALSE, bw = bandwidth_nrd_OC) +scale_fill_identity() +labs(title ="Oceanian SDG goals boxplot", x ="Goals", y ="Score") +geom_boxplot(width =0.1, outlier.size =1, fill ='white') +scale_y_continuous(labels = scales::label_number()) +theme_classic() +theme(axis.text.x =element_text(angle =90, vjust =0.5, hjust =1))grid.arrange(AF,EU,AS,AM,OC, ncol =2, nrow =3)# Correlation_goals <- melt(Correlation_overall[,2:18])# ggplot(Correlation_goals, aes(x= variable, y= value)) + # geom_violin(trim=FALSE, fill="orange") +# labs(title="Merged goals violin boxplot",x="Goals", y = "Distribution") +# geom_boxplot(width=0.1, outlier.size = 1) +# scale_y_continuous(labels = scales::label_number()) + #limits = c(0, 100)# theme_classic() +# theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust=1))#### WHY GOING BELOW 0 TO > 100 ?? SCORES ONLY FROM 0 TO 100
# Step 1: Combine all data into one data frame with a region identifier# Melt the datadata_long <- data_question1 %>%select(continent, overallscore, goal1, goal2, goal3, goal4, goal5, goal6, goal7, goal8, goal9, goal10, goal11, goal12, goal13, goal15, goal16, goal17) %>%melt()# Calculate medians and colorsmedians <- data_long %>%group_by(variable) %>%summarize(median_value =median(value), .groups ='drop')medians$color <-ifelse(medians$median_value >75, "lightblue", ifelse(medians$median_value <25, "red", 'orange'))# Join the medians back to the long datadata_long <-left_join(data_long, medians, by ="variable")# Calculate the bandwidthbandwidth_nrd <-bw.nrd(data_long$value)# Create the plotp <-ggplot(data_long, aes(x = variable, y = value, fill = color)) +geom_violin(trim =FALSE, bw = bandwidth_nrd) +geom_boxplot(width =0.1, outlier.size =1, fill ='white') +scale_fill_identity() +labs(title ="SDG Goals by Region", x ="Goals", y ="Score") +facet_grid(continent ~ ., scales ="free_y") +scale_y_continuous(labels = scales::label_number()) +theme_classic() +theme(axis.text.x =element_text(angle =90, vjust =0.5, hjust=1))# Print the plotprint(p)
>>>>>>> Stashed changes
We see different schemes among the different goals. Indeed some are quite homogeneous with a small spread of values (e.g. overall score, goals 2 and 8) while others have a large spread of values (e.g. goals 1 and 10). Goals 1, 3, 4, 7, 9, 10 and 13 have values across all possible percentages. Goals 2, 5, 8, 13 and 17 have extreme values situated outside the 95% confidence interval. It is interesting to see that goal 8 (decent work and economic growth) is the one with smaller spread of values, whereas goal 1 (no poverty) have the higher distance between the first and the third quartile. Goal 2 (no hunger) has a tight spread of values, but with the greater amount of outliers in the smaller values, meaning hunger is similar across most countries, but when it differs it is in very bad manner.
We now display boxplpots for the different variables of the human freedom index, and then also for our other independent variables.
<<<<<<< Updated upstream
Code
#for Human Freedom Index scores #Africa data_Q1_Africa_HFI_long <-melt(data_Q1_Africa[,29:40])medians_HFI_AF <- data_Q1_Africa_HFI_long %>%group_by(variable) %>%summarize(median_value =median(value))medians_HFI_AF$color <-ifelse(medians_HFI_AF$median_value >7.5, "lightblue", ifelse(medians_HFI_AF$median_value <2.5, "red", 'orange'))data_Q1_Africa_HFI_long <- data_Q1_Africa_HFI_long %>%left_join(medians_HFI_AF, by ="variable")bandwidth_nrd_HFI_AF <-bw.nrd(data_Q1_Africa_HFI_long$value)HFI_AF <-ggplot(data_Q1_Africa_HFI_long, aes(x = variable, y = value, fill = color)) +geom_violin(trim =FALSE, bw = bandwidth_nrd_HFI_AF) +scale_fill_identity() +labs(title ="African HFI Scores boxplot", x ="Human Freedom Index goals", y ="Score") +geom_boxplot(width =0.1, outlier.size =1, fill ='white') +scale_y_continuous(labels = scales::label_number()) +theme_classic() +theme(axis.text.x =element_text(angle =90, vjust =0.5, hjust =1))#Europedata_Q1_Europe_HFI_long <-melt(data_Q1_Europe[,29:40])medians_HFI_EU <- data_Q1_Europe_HFI_long %>%group_by(variable) %>%summarize(median_value =median(value))medians_HFI_EU$color <-ifelse(medians_HFI_EU$median_value >7.5, "lightblue", ifelse(medians_HFI_EU$median_value <2.5, "red", 'orange'))data_Q1_Europe_HFI_long <- data_Q1_Europe_HFI_long %>%left_join(medians_HFI_EU, by ="variable")bandwidth_nrd_HFI_EU <-bw.nrd(data_Q1_Europe_HFI_long$value)HFI_EU <-ggplot(data_Q1_Europe_HFI_long, aes(x = variable, y = value, fill = color)) +geom_violin(trim =FALSE, bw = bandwidth_nrd_HFI_EU) +scale_fill_identity() +labs(title ="Europe HFI Scores boxplot", x ="Human Freedom Index goals", y ="Score") +geom_boxplot(width =0.1, outlier.size =1, fill ='white') +scale_y_continuous(labels = scales::label_number()) +theme_classic() +theme(axis.text.x =element_text(angle =90, vjust =0.5, hjust =1))#Asiadata_Q1_Asia_HFI_long <-melt(data_Q1_Asia[,29:40])medians_HFI_AS <- data_Q1_Asia_HFI_long %>%group_by(variable) %>%summarize(median_value =median(value))medians_HFI_AS$color <-ifelse(medians_HFI_AS$median_value >7.5, "lightblue", ifelse(medians_HFI_AS$median_value <2.5, "red", 'orange'))data_Q1_Asia_HFI_long <- data_Q1_Asia_HFI_long %>%left_join(medians_HFI_AS, by ="variable")bandwidth_nrd_HFI_AS <-bw.nrd(data_Q1_Asia_HFI_long$value)HFI_AS <-ggplot(data_Q1_Asia_HFI_long, aes(x = variable, y = value, fill = color)) +geom_violin(trim =FALSE, bw = bandwidth_nrd_HFI_AS) +scale_fill_identity() +labs(title ="Asian HFI Scores boxplot", x ="Human Freedom Index goals", y ="Score") +geom_boxplot(width =0.1, outlier.size =1, fill ='white') +scale_y_continuous(labels = scales::label_number()) +theme_classic() +theme(axis.text.x =element_text(angle =90, vjust =0.5, hjust =1))#Americadata_Q1_America_HFI_long <-melt(data_Q1_Americas[,29:40])medians_HFI_AM <- data_Q1_America_HFI_long %>%group_by(variable) %>%summarize(median_value =median(value))medians_HFI_AM$color <-ifelse(medians_HFI_AM$median_value >7.5, "lightblue", ifelse(medians_HFI_AM$median_value <2.5, "red", 'orange'))data_Q1_America_HFI_long <- data_Q1_America_HFI_long %>%left_join(medians_HFI_AM, by ="variable")bandwidth_nrd_HFI_AM <-bw.nrd(data_Q1_America_HFI_long$value)HFI_AM <-ggplot(data_Q1_Asia_HFI_long, aes(x = variable, y = value, fill = color)) +geom_violin(trim =FALSE, bw = bandwidth_nrd_HFI_AM) +scale_fill_identity() +labs(title ="America HFI Scores boxplot", x ="Human Freedom Index goals", y ="Score") +geom_boxplot(width =0.1, outlier.size =1, fill ='white') +scale_y_continuous(labels = scales::label_number()) +theme_classic() +theme(axis.text.x =element_text(angle =90, vjust =0.5, hjust =1))#Oceaniadata_Q1_Oceania_HFI_long <-melt(data_Q1_Oceania[,29:40])medians_HFI_OC <- data_Q1_Oceania_HFI_long %>%group_by(variable) %>%summarize(median_value =median(value))medians_HFI_OC$color <-ifelse(medians_HFI_OC$median_value >7.5, "lightblue", ifelse(medians_HFI_OC$median_value <2.5, "red", 'orange'))data_Q1_Oceania_HFI_long <- data_Q1_Oceania_HFI_long %>%left_join(medians_HFI_OC, by ="variable")bandwidth_nrd_HFI_OC <-bw.nrd(data_Q1_Oceania_HFI_long$value)HFI_OC <-ggplot(data_Q1_Oceania_HFI_long, aes(x = variable, y = value, fill = color)) +geom_violin(trim =FALSE, bw = bandwidth_nrd_HFI_OC) +scale_fill_identity() +labs(title ="Oceanian HFI Scores boxplot", x ="Human Freedom Index goals", y ="Score") +geom_boxplot(width =0.1, outlier.size =1, fill ='white') +scale_y_continuous(labels = scales::label_number()) +theme_classic() +theme(axis.text.x =element_text(angle =90, vjust =0.5, hjust =1))grid.arrange(HFI_AF,HFI_EU,HFI_AS,HFI_AM,HFI_OC, ncol =2, nrow =3)# boxplot(Correlation_overall[23:34], # las = 2, # Makes the axis labels perpendicular to the axis# par(mar = c(7, 5, 2, 1)), # Adjusts the margins to fit all labels# cex.axis = 0.7, # Reduces the size of the axis labels# cex.lab = 1, # Reduces the size of the x and y labels# notch = TRUE, # Specifies whether to add notches or not# main = "Merged Human Freedom Index scores boxplot", # ylab = "Score") # Y-axis label# Correlation_HFI <- melt(Correlation_overall[,23:34])# ggplot(Correlation_HFI, aes(x= variable, y= value)) + # geom_violin(trim=FALSE, fill="orange")+# labs(title="Merged Human Freedom Index scores violin boxplot",x="Variables", y = "Score")+# geom_boxplot(width=0.1, outlier.size = 1)+# scale_y_continuous(labels = scales::label_number()) + #limits = c(0, 100)# theme_classic() +# theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust=1))v1 <-ggplot(Correlation_overall, aes(x=factor(1), y= GDPpercapita)) +geom_violin(trim=FALSE, fill="orange")+labs(title="Violin plot of GDP per capita",x="GDP per capita", y ="Distribution")+geom_boxplot(width=0.1, outlier.size =1)+scale_y_continuous(labels = scales::label_number()) +# Format y-axis labelstheme_classic()v2 <-ggplot(Correlation_overall, aes(x=factor(1), y= unemployment.rate)) +geom_violin(trim=FALSE, fill="orange")+labs(title="Violin plot of unemployment rate",x="Unemployment rate", y ="Distribution")+geom_boxplot(width=0.1, outlier.size =1)+scale_y_continuous(labels = scales::label_number()) +# Format y-axis labelstheme_classic()v3 <-ggplot(Correlation_overall, aes(x=factor(1), y= MilitaryExpenditurePercentGDP)) +geom_violin(trim=FALSE, fill="orange")+labs(title="Violin plot of military expenditure by percentage of GDP",x="Military Expenditure", y ="Distribution")+geom_boxplot(width=0.1, outlier.size =1)+scale_y_continuous(labels = scales::label_number()) +# Format y-axis labelstheme_classic()v4 <-ggplot(Correlation_overall, aes(x=factor(1), y= internet_usage)) +geom_violin(trim=FALSE, fill="orange")+labs(title="Violin plot of internet_usage",x="internet_usage", y ="Distribution")+geom_boxplot(width=0.1, outlier.size =1)+scale_y_continuous(labels = scales::label_number()) +# Format y-axis labelstheme_classic()grid.arrange(v1,v2,v3,v4, ncol =2, nrow =2)
=======
Code
#for Human Freedom Index scores #Africa data_Q1_Africa_HFI_long <-melt(data_Q1_Africa[,29:40])medians_HFI_AF <- data_Q1_Africa_HFI_long %>%group_by(variable) %>%summarize(median_value =median(value))medians_HFI_AF$color <-ifelse(medians_HFI_AF$median_value >7.5, "lightblue", ifelse(medians_HFI_AF$median_value <2.5, "red", 'orange'))data_Q1_Africa_HFI_long <- data_Q1_Africa_HFI_long %>%left_join(medians_HFI_AF, by ="variable")bandwidth_nrd_HFI_AF <-bw.nrd(data_Q1_Africa_HFI_long$value)HFI_AF <-ggplot(data_Q1_Africa_HFI_long, aes(x = variable, y = value, fill = color)) +geom_violin(trim =FALSE, bw = bandwidth_nrd_HFI_AF) +scale_fill_identity() +labs(title ="African HFI Scores boxplot", x ="Human Freedom Index goals", y ="Score") +geom_boxplot(width =0.1, outlier.size =1, fill ='white') +scale_y_continuous(labels = scales::label_number()) +theme_classic() +theme(axis.text.x =element_text(angle =90, vjust =0.5, hjust =1))#Europedata_Q1_Europe_HFI_long <-melt(data_Q1_Europe[,29:40])medians_HFI_EU <- data_Q1_Europe_HFI_long %>%group_by(variable) %>%summarize(median_value =median(value))medians_HFI_EU$color <-ifelse(medians_HFI_EU$median_value >7.5, "lightblue", ifelse(medians_HFI_EU$median_value <2.5, "red", 'orange'))data_Q1_Europe_HFI_long <- data_Q1_Europe_HFI_long %>%left_join(medians_HFI_EU, by ="variable")bandwidth_nrd_HFI_EU <-bw.nrd(data_Q1_Europe_HFI_long$value)HFI_EU <-ggplot(data_Q1_Europe_HFI_long, aes(x = variable, y = value, fill = color)) +geom_violin(trim =FALSE, bw = bandwidth_nrd_HFI_EU) +scale_fill_identity() +labs(title ="Europe HFI Scores boxplot", x ="Human Freedom Index goals", y ="Score") +geom_boxplot(width =0.1, outlier.size =1, fill ='white') +scale_y_continuous(labels = scales::label_number()) +theme_classic() +theme(axis.text.x =element_text(angle =90, vjust =0.5, hjust =1))#Asiadata_Q1_Asia_HFI_long <-melt(data_Q1_Asia[,29:40])medians_HFI_AS <- data_Q1_Asia_HFI_long %>%group_by(variable) %>%summarize(median_value =median(value))medians_HFI_AS$color <-ifelse(medians_HFI_AS$median_value >7.5, "lightblue", ifelse(medians_HFI_AS$median_value <2.5, "red", 'orange'))data_Q1_Asia_HFI_long <- data_Q1_Asia_HFI_long %>%left_join(medians_HFI_AS, by ="variable")bandwidth_nrd_HFI_AS <-bw.nrd(data_Q1_Asia_HFI_long$value)HFI_AS <-ggplot(data_Q1_Asia_HFI_long, aes(x = variable, y = value, fill = color)) +geom_violin(trim =FALSE, bw = bandwidth_nrd_HFI_AS) +scale_fill_identity() +labs(title ="Asian HFI Scores boxplot", x ="Human Freedom Index goals", y ="Score") +geom_boxplot(width =0.1, outlier.size =1, fill ='white') +scale_y_continuous(labels = scales::label_number()) +theme_classic() +theme(axis.text.x =element_text(angle =90, vjust =0.5, hjust =1))#Americadata_Q1_America_HFI_long <-melt(data_Q1_Americas[,29:40])medians_HFI_AM <- data_Q1_America_HFI_long %>%group_by(variable) %>%summarize(median_value =median(value))medians_HFI_AM$color <-ifelse(medians_HFI_AM$median_value >7.5, "lightblue", ifelse(medians_HFI_AM$median_value <2.5, "red", 'orange'))data_Q1_America_HFI_long <- data_Q1_America_HFI_long %>%left_join(medians_HFI_AM, by ="variable")bandwidth_nrd_HFI_AM <-bw.nrd(data_Q1_America_HFI_long$value)HFI_AM <-ggplot(data_Q1_Asia_HFI_long, aes(x = variable, y = value, fill = color)) +geom_violin(trim =FALSE, bw = bandwidth_nrd_HFI_AM) +scale_fill_identity() +labs(title ="America HFI Scores boxplot", x ="Human Freedom Index goals", y ="Score") +geom_boxplot(width =0.1, outlier.size =1, fill ='white') +scale_y_continuous(labels = scales::label_number()) +theme_classic() +theme(axis.text.x =element_text(angle =90, vjust =0.5, hjust =1))#Oceaniadata_Q1_Oceania_HFI_long <-melt(data_Q1_Oceania[,29:40])medians_HFI_OC <- data_Q1_Oceania_HFI_long %>%group_by(variable) %>%summarize(median_value =median(value))medians_HFI_OC$color <-ifelse(medians_HFI_OC$median_value >7.5, "lightblue", ifelse(medians_HFI_OC$median_value <2.5, "red", 'orange'))data_Q1_Oceania_HFI_long <- data_Q1_Oceania_HFI_long %>%left_join(medians_HFI_OC, by ="variable")bandwidth_nrd_HFI_OC <-bw.nrd(data_Q1_Oceania_HFI_long$value)HFI_OC <-ggplot(data_Q1_Oceania_HFI_long, aes(x = variable, y = value, fill = color)) +geom_violin(trim =FALSE, bw = bandwidth_nrd_HFI_OC) +scale_fill_identity() +labs(title ="Oceanian HFI Scores boxplot", x ="Human Freedom Index goals", y ="Score") +geom_boxplot(width =0.1, outlier.size =1, fill ='white') +scale_y_continuous(labels = scales::label_number()) +theme_classic() +theme(axis.text.x =element_text(angle =90, vjust =0.5, hjust =1))grid.arrange(HFI_AF,HFI_EU,HFI_AS,HFI_AM,HFI_OC, ncol =2, nrow =3)# boxplot(Correlation_overall[23:34], # las = 2, # Makes the axis labels perpendicular to the axis# par(mar = c(7, 5, 2, 1)), # Adjusts the margins to fit all labels# cex.axis = 0.7, # Reduces the size of the axis labels# cex.lab = 1, # Reduces the size of the x and y labels# notch = TRUE, # Specifies whether to add notches or not# main = "Merged Human Freedom Index scores boxplot", # ylab = "Score") # Y-axis label# Correlation_HFI <- melt(Correlation_overall[,23:34])# ggplot(Correlation_HFI, aes(x= variable, y= value)) + # geom_violin(trim=FALSE, fill="orange")+# labs(title="Merged Human Freedom Index scores violin boxplot",x="Variables", y = "Score")+# geom_boxplot(width=0.1, outlier.size = 1)+# scale_y_continuous(labels = scales::label_number()) + #limits = c(0, 100)# theme_classic() +# theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust=1))v1 <-ggplot(Correlation_overall, aes(x=factor(1), y= GDPpercapita)) +geom_violin(trim=FALSE, fill="orange")+labs(title="Violin plot of GDP per capita",x="GDP per capita", y ="Distribution")+geom_boxplot(width=0.1, outlier.size =1)+scale_y_continuous(labels = scales::label_number()) +# Format y-axis labelstheme_classic()v2 <-ggplot(Correlation_overall, aes(x=factor(1), y= unemployment.rate)) +geom_violin(trim=FALSE, fill="orange")+labs(title="Violin plot of unemployment rate",x="Unemployment rate", y ="Distribution")+geom_boxplot(width=0.1, outlier.size =1)+scale_y_continuous(labels = scales::label_number()) +# Format y-axis labelstheme_classic()v3 <-ggplot(Correlation_overall, aes(x=factor(1), y= MilitaryExpenditurePercentGDP)) +geom_violin(trim=FALSE, fill="orange")+labs(title="Violin plot of military expenditure by percentage of GDP",x="Military Expenditure", y ="Distribution")+geom_boxplot(width=0.1, outlier.size =1)+scale_y_continuous(labels = scales::label_number()) +# Format y-axis labelstheme_classic()v4 <-ggplot(Correlation_overall, aes(x=factor(1), y= internet_usage)) +geom_violin(trim=FALSE, fill="orange")+labs(title="Violin plot of internet_usage",x="internet_usage", y ="Distribution")+geom_boxplot(width=0.1, outlier.size =1)+scale_y_continuous(labels = scales::label_number()) +# Format y-axis labelstheme_classic()grid.arrange(v1,v2,v3,v4, ncol =2, nrow =2)
>>>>>>> Stashed changes
We now look at the variables in a summary table to have a more precise view of the numbers.
X
code
year
country
continent
region
overallscore
goal1
goal2
goal3
goal4
goal5
goal6
goal7
goal8
goal9
goal10
goal11
goal12
goal13
goal15
goal16
goal17
Min. : 1
Length:3565
Min. :2000
Length:3565
Length:3565
Length:3565
Min. :37.4
Min. : 0.0
Min. :16.5
Min. : 5.9
Min. : 0.0
Min. : 3.5
Min. :23.3
Min. : 0.1
Min. :40.0
Min. : 0.3
Min. : 0.0
Min. :20.3
Min. :32.9
Min. : 0.0
Min. :26.0
Min. :27.9
Min. :15.1
1st Qu.: 892
Class :character
1st Qu.:2005
Class :character
Class :character
Class :character
1st Qu.:55.0
1st Qu.: 44.5
1st Qu.:52.6
1st Qu.:44.3
1st Qu.: 55.6
1st Qu.:43.2
1st Qu.:53.0
1st Qu.:41.5
1st Qu.:64.0
1st Qu.:15.5
1st Qu.: 35.2
1st Qu.:55.8
1st Qu.:67.9
1st Qu.:72.9
1st Qu.:55.0
1st Qu.:51.5
1st Qu.:46.1
Median :1783
Mode :character
Median :2011
Mode :character
Mode :character
Mode :character
Median :65.5
Median : 87.4
Median :58.9
Median :70.9
Median : 80.6
Median :58.0
Median :65.3
Median :65.5
Median :70.2
Median :29.4
Median : 62.2
Median :75.3
Median :84.6
Median :90.8
Median :65.1
Median :61.4
Median :55.4
Mean :1783
NA
Mean :2011
NA
NA
NA
Mean :64.0
Mean : 71.7
Mean :58.0
Mean :64.1
Mean : 72.0
Mean :56.0
Mean :65.0
Mean :57.9
Mean :70.0
Mean :37.5
Mean : 58.3
Mean :70.3
Mean :79.3
Mean :82.1
Mean :65.0
Mean :62.6
Mean :55.7
3rd Qu.:2674
NA
3rd Qu.:2017
NA
NA
NA
3rd Qu.:72.4
3rd Qu.: 98.8
3rd Qu.:65.3
3rd Qu.:81.4
3rd Qu.: 94.5
3rd Qu.:68.9
3rd Qu.:75.2
3rd Qu.:72.6
3rd Qu.:76.6
3rd Qu.:53.9
3rd Qu.: 81.6
3rd Qu.:85.1
3rd Qu.:94.1
3rd Qu.:97.2
3rd Qu.:74.3
3rd Qu.:74.6
3rd Qu.:65.1
Max. :3565
NA
Max. :2022
NA
NA
NA
Max. :86.8
Max. :100.0
Max. :83.4
Max. :97.3
Max. :100.0
Max. :94.0
Max. :95.1
Max. :99.6
Max. :88.7
Max. :99.2
Max. :100.0
Max. :99.1
Max. :99.0
Max. :99.9
Max. :97.9
Max. :96.0
Max. :96.8
NA
NA
NA
NA
NA
NA
NA
NA's :276
NA
NA
NA
NA
NA
NA
NA
NA
NA's :276
NA
NA
NA
NA
NA
NA
3.2 Focus on the influence of the factors over the SDG scores
<<<<<<< Updated upstream
Using our cleaned dataset, we first want to observe how each of our variables are correlated with the others. For that, we will use a heatmap. Given that most of our variables are not normally distributed, we will use the Spearman method to calculate the correlation.
Code
#### Correlations between variables Heatmap ####Correlation_overall <-data_question1 %>%# selection of the numerical dataselect(population:ef_regulation)cor_matrix_sper <-# calculation of the correlation matrixcor(Correlation_overall, method ="spearman", use ="everything")cor_melted <-# wide to long transformationmelt(cor_matrix_sper)ggplot(data = cor_melted, aes(Var1, Var2, fill = value)) +geom_tile() +scale_fill_gradient2(low ="blue", high ="red", mid ="white", midpoint =0, limit =c(-1, 1), space ="Lab", name="Spearman\nCorrelation") +theme_minimal() +theme(axis.text.x =element_text(angle =45, vjust =1, size =8, hjust =1),axis.text.y =element_text(size =8)) +coord_fixed() +labs(x ='', y ='', title ='Correlation Matrix Heatmap')#do 3 different heatmaps : goals on goals, goals on other variables except goals, variables on variables (except goals)
In the correlation matrix heatmap, we observe that goals 1 to 11 are predominantly positively correlated. Conversely, goals 12 and 13 exhibit negative correlations with most variables, except between themselves where they are strongly correlated. Additionally, there is a notable strong correlation among personal freedom variables (pf), reflecting scores from the Human Freedom Index on movement, religion, assembly, and expression.
In order to have an overview of the relationship between our independent variables and the SDG overall score, we make several graphs containing the Spearman correlation coefficient between the variable, the scatter plots describing the relationship between the variables, as well as the distribution of each variable.
Code
#### Spearman's correlation coeff ####panel.hist <-function(x, ...){ usr <-par("usr"); on.exit(par(usr)) par(usr =c(usr[1:2], 0, 1.5) ) h <-hist(x, plot =FALSE) breaks <- h$breaks; nB <-length(breaks) y <- h$counts; y <- y/max(y) rect(breaks[-nB], 0, breaks[-1], y, col ="lightgreen", ...)}panel.cor <-function(x, y, digits =2, prefix ="", cex.cor, ...){ usr <-par("usr"); on.exit(par(usr)) par(usr =c(0, 1, 0, 1)) r <-cor(x, y, method ="spearman") txt <-format(c(r, 0.123456789), digits = digits)[1] txt <-paste0(prefix, txt) if(missing(cex.cor)) cex.cor <-0.8/strwidth(txt) text(0.5, 0.5, txt, cex = cex.cor * r)}# # Independent variables pairs(Correlation_overall[,c("overallscore", "unemployment.rate", "GDPpercapita", "MilitaryExpenditurePercentGDP", "internet_usage")], upper.panel=panel.cor, diag.panel=panel.hist, main="Correlation table and distribution of various variables")
=======
After importing our our cleaned data, we looked first at the correlations between our numerical variables.
Code
#### Correlations between variables ####sdg_scores2 <- data_question1[, c('goal1', 'goal2', 'goal3', 'goal4', 'goal5', 'goal6','goal7', 'goal8', 'goal9', 'goal10', 'goal11', 'goal12','goal13', 'goal15', 'goal16', 'goal17')]Correlation_overall <- data_question1 %>%select(population:ef_regulation)#before computing Pearson -> logarithmic transformation may be required for some variables with high skewness# Calculating skewness for each variableCorrelation_overall_skew <- Correlation_overallCorrelation_overall_sqrt <- Correlation_overallskewness_values <-sapply(Correlation_overall_skew, e1071::skewness)# Identifying highly skewed variableshighly_skewed_vars <-names(skewness_values[abs(skewness_values) >1])highly_skewed_vars_sqrt <-names(skewness_values[abs(skewness_values) >1])# Applying logarithmic transformationCorrelation_overall_skew[highly_skewed_vars] <-lapply(Correlation_overall_skew[highly_skewed_vars], function(x) log1p(x))#applying squart root transformationCorrelation_overall_sqrt[highly_skewed_vars_sqrt] <-lapply(Correlation_overall_sqrt[highly_skewed_vars_sqrt], function(x) sqrt(x))new_skewness_values <-sapply(Correlation_overall_skew[highly_skewed_vars], e1071::skewness)new_skewness_values_sqrt <-sapply(Correlation_overall_sqrt[highly_skewed_vars_sqrt], e1071::skewness)#après transformation, il reste toujours beaucoup de variables avec un high skewness values print(new_skewness_values), print(new_skewness_values_sqrt)cor_matrix_log <-cor(Correlation_overall_skew, use ="everything")cor_matrix_sqrt <-cor(Correlation_overall_sqrt, use ="everything")cor_matrix_sper <-cor(Correlation_overall, method ="spearman", use ="everything")datatable(cor_matrix_log, options =list(pageLength =10, class ="hover", searchHighlight =TRUE,columnDefs =list(list(targets ="_all",render =JS("function(data, type, row, meta){"," if(type === 'display'){"," return parseFloat(data).toFixed(2)"," }"," return data;","}")))),rownames =FALSE)
By doing so, we obtain a lot of positive and negative correlations. To help us to better understand and having a overall vision of the situation, we used the following heatmap.
Code
#### Heatmap ####cor_melted <-melt(cor_matrix_sper)ggplot(data = cor_melted, aes(Var1, Var2, fill = value)) +geom_tile() +scale_fill_gradient2(low ="blue", high ="red", mid ="white", midpoint =0, limit =c(-1, 1), space ="Lab", name="Spearman\nCorrelation") +theme_minimal() +theme(axis.text.x =element_text(angle =45, vjust =1, size =8, hjust =1),axis.text.y =element_text(size =8)) +coord_fixed() +labs(x ='', y ='', title ='Correlation Matrix Heatmap')#do 3 different heatmaps : goals on goals, goals on other variables except goals, variables on variables (except goals)
>>>>>>> Stashed changes
<<<<<<< Updated upstream
The overall SDG achievement score is highly correlated with the percentage of people using the internet (r=.79) and GDP per capita (r=.60). The unemployement rate as well as the military expenditures in percentage of GDP per capita do not seem to play a role. However, this is only for the overall score.
The overall SDG achievement score is highly correlated with “personal freedom: law” (p=.69) and “personal freedom: identity” (p=.62). The other dimensions of personal freedom do not seem to have important influence. Regarding the distribution of the personal freedom variables, we notice that except for law, all have right-skewed distributions meaning that most of the countries have high scores.
The overall SDG achievement score is highly correlated with “economical freedom: legal” (p=.77), “economical trade: legal” (p=.67) and “economical freedom: money” (p=.6), while the other dimensions of economic freedom do not seem to have important influence. Regarding the distribution of the economic freedom variables, we notice more heterogeneous distributions and scores across the various countries than for personal freedom.
3.2.1 Looking at SDGs
ADD GREEN GRAPH HERE TO HAVE ALSO THE DISTRIBUTION
As we can see in the graph, most of our goals are correlated toghether. We will nown perform a PCA analysis to see how our variables are explained. You can see below the Scree plot of our PCA analysis.
As we can see, Dimension 1 explain already more than 60% of the variations in our data. With Dimension 2, it goes up to around 70%. We can now plot our data with our two firsts dimensions.
Concerning the SDG goals, we conclude that most of our variables are going along the 1st component, except the goals 10 and 15 that are rather uncorrelated with the dimension 1. In addition, as seen before, the goals 12 and 13 are negatively correlated to the other goals. With a eigenvalue bigger than 1 for the first two components, we conclude that there are only 2 dimensions to take into account, according to the Kaiser-Guttman’s rule. Nevertheless, they are explaining less than 80% of cumulated variance.
Now concerning the Human Freedom Index scores, most of the variables are positively correlated to the dimension 1, slightly less for the PF religion and security, and finaly the EF government variable is uncorrelated to the dimension 1. With a eigenvalue bigger than 1 for the three first components, we conclude that there are 3 dimensions to take into account. Nevertheless, again, they are explaining less than 80% of cumulated variance.
Code
#### Kmean clustering ####data1_scaled <-scale(Correlation_overall)rownames(data1_scaled) <-seq_along(row.names(data1_scaled))fviz_nbclust(data1_scaled, kmeans, method="wss")kmean <-kmeans(data1_scaled, 7, nstart =25)fviz_cluster(kmean, data=data1_scaled, repel=FALSE, depth =NULL, ellipse.type ="norm", labelsize =0, pointsize =0.5)### NOW CLUSTERING BY COUNTRY? AND TAKE MEAN OF EVERY VARIABLE ON EVERY CONCERNED YEAR?
Due to the large number of data, the visualization of the clusters using the kmean method is not really relevant. In addition, by clustering our data, we are trying to get group that differ from eachother but with little variation of the observations within the same cluster. Here, only 60.6% of the variance is explained by the variation between clusters. This is not enough.
=======
In the correlation matrix heatmap, we can notice that many goals from 1 to 11 are actually positively correlated together. On another hand, the goals 12 and 13 are have negative relationships with the majority of our variables, except between themself, whereas they are strongely correlated. In addition, we can notice another strongly correlation between personal freedom variables (pf) related to the scores given by the Human Freedom Index on movement, religion, assembly and expression.
In order to have an overview of the relationship between our independent variables and the SDG overall score, we make several graphs containing the Pearson correlation coefficient between the variable, the scatter plots describing the relationship between the variables, as well as the distribution of each variable.
Code
#### Spearman's correlation coeff ####panel.hist <-function(x, ...){ usr <-par("usr"); on.exit(par(usr)) par(usr =c(usr[1:2], 0, 1.5) ) h <-hist(x, plot =FALSE) breaks <- h$breaks; nB <-length(breaks) y <- h$counts; y <- y/max(y) rect(breaks[-nB], 0, breaks[-1], y, col ="lightgreen", ...)}panel.cor <-function(x, y, digits =2, prefix ="", cex.cor, ...){ usr <-par("usr"); on.exit(par(usr)) par(usr =c(0, 1, 0, 1)) r <-cor(x, y, method ="spearman") txt <-format(c(r, 0.123456789), digits = digits)[1] txt <-paste0(prefix, txt) if(missing(cex.cor)) cex.cor <-0.8/strwidth(txt) text(0.5, 0.5, txt, cex = cex.cor * r)}# # Independent variables # #with log transformation# pairs(Correlation_overall_skew[,c("overallscore", "unemployment.rate", "GDPpercapita", "MilitaryExpenditurePercentGDP", "internet_usage")], upper.panel=panel.cor, diag.panel=panel.hist, main="Correlation table and distribution of various variables")# #with square root transformation# pairs(Correlation_overall_sqrt[,c("overallscore", "unemployment.rate", "GDPpercapita", "MilitaryExpenditurePercentGDP", "internet_usage")], upper.panel=panel.cor, diag.panel=panel.hist, main="Correlation table and distribution of various variables")
The overall SDG achievement score is highly correlated with the percentage of people using the internet (r=.79) and GDP per capita (r=.60). The unemployement rate as well as the military expenditures in percentage of GDP per capita do not seem to play a role. However, this is only for the overall score.
Code
#with log transformation# pairs(Correlation_overall_skew[,c("overallscore", "pf_law", "pf_security", "pf_movement", "pf_religion", "pf_assembly", "pf_expression", "pf_identity")], upper.panel=panel.cor, diag.panel=panel.hist, main="Correlation table and distribution of personal freedom variables")# # #with square root transformation# pairs(Correlation_overall_sqrt[,c("overallscore", "pf_law", "pf_security", "pf_movement", "pf_religion", "pf_assembly", "pf_expression", "pf_identity")], upper.panel=panel.cor, diag.panel=panel.hist, main="Correlation table and distribution of personal freedom variables")
The overall SDG achievement score is highly correlated with “personal freedom: law” (p=.69) and “personal freedom: identity” (p=.62). The other dimensions of personal freedom do not seem to have important influence. Regarding the distribution of the personal freedom variables, we notice that except for law, all have right-skewed distributions meaning that most of the countries have high scores.
Code
#with log transformationpairs(Correlation_overall_skew[,c("overallscore", "ef_government", "ef_legal", "ef_money", "ef_trade", "ef_regulation")], upper.panel=panel.cor, diag.panel=panel.hist, main="Correlation table and distribution of economic freedom variables")#with square root transformationpairs(Correlation_overall_sqrt[,c("overallscore", "ef_government", "ef_legal", "ef_money", "ef_trade", "ef_regulation")], upper.panel=panel.cor, diag.panel=panel.hist, main="Correlation table and distribution of economic freedom variables")
The overall SDG achievement score is highly correlated with “economical freedom: legal” (p=.77), “economical trade: legal” (p=.67) and “economical freedom: money” (p=.6), while the other dimensions of economic freedom do not seem to have important influence. Regarding the distribution of the economic freedom variables, we notice more heterogeneous distributions and scores across the various countries than for personal freedom.
>>>>>>> Stashed changes
3.3 Focus on the influence of events over the SDG scores
In order to have an overview of the relationship between the different events variables and the SDG overall score, we make several graphs containing the Pearson correlation coefficient between the variable, the scatter plots describing the relationship between the variables, as well as the distribution of each variable.
<<<<<<< Updated upstream
Code
lower.panel <-function(x, y, ...){points(x, y, pch =20, col ="black", cex =0.2)}evaluateCorrelationStars <-function(correlation) {if (abs(correlation) >=0.7) {return("*****") # Strong correlation: 5 stars } elseif (abs(correlation) >=0.5) {return("****") # Moderate correlation: 4 stars } elseif (abs(correlation) >=0.3) {return("***") # Fair correlation: 3 stars } elseif (abs(correlation) >=0.1) {return("**") # Weak correlation: 2 stars } else {return("*") # Very weak correlation: 1 star }}# panel.cor function with stars alongside correlation coefficientspanel.cor_stars <-function(x, y, digits =2, prefix ="", cex.cor, ...) { usr <-par("usr"); on.exit(par(usr)) par(usr =c(0, 1, 0, 1)) r <-cor(x, y) stars <-evaluateCorrelationStars(r) txt <-paste0(format(c(r, 0.123456789), digits = digits)[1], " ", stars)if(missing(cex.cor)) cex.cor <-0.5/strwidth(txt)text(0.5, 0.5, txt, cex = cex.cor)}pairs(data_question3_1[, c("overallscore", "total_affected", "total_deaths")], upper.panel = panel.cor_stars,diag.panel = panel.hist,lower.panel = lower.panel, main ="Correlation table and distribution of Disaster variables")
=======
Code
lower.panel <-function(x, y, ...){points(x, y, pch =20, col ="black", cex =0.2)}evaluateCorrelationStars <-function(correlation) {if (abs(correlation) >=0.7) {return("*****") # Strong correlation: 5 stars } elseif (abs(correlation) >=0.5) {return("****") # Moderate correlation: 4 stars } elseif (abs(correlation) >=0.3) {return("***") # Fair correlation: 3 stars } elseif (abs(correlation) >=0.1) {return("**") # Weak correlation: 2 stars } else {return("*") # Very weak correlation: 1 star }}# panel.cor function with stars alongside correlation coefficientspanel.cor_stars <-function(x, y, digits =2, prefix ="", cex.cor, ...) { usr <-par("usr"); on.exit(par(usr)) par(usr =c(0, 1, 0, 1)) r <-cor(x, y) stars <-evaluateCorrelationStars(r) txt <-paste0(format(c(r, 0.123456789), digits = digits)[1], " ", stars)if(missing(cex.cor)) cex.cor <-0.5/strwidth(txt)text(0.5, 0.5, txt, cex = cex.cor)}pairs(data_question3_1[, c("overallscore", "total_affected", "total_deaths")], upper.panel = panel.cor_stars,diag.panel = panel.hist,lower.panel = lower.panel, main ="Correlation table and distribution of Disaster variables")
>>>>>>> Stashed changes
The different variables used to materialize the impact of climate disasters do not seem to have important influence on the overall score, but we will further explore for the different SDGs, since we believe that such disasters have a specific influence on some SDGs.
<<<<<<< Updated upstream
Code
lower.panel <-function(x, y, ...){points(x, y, pch =20, col ="black", cex =0.2)}evaluateCorrelationStars <-function(correlation) {if (abs(correlation) >=0.7) {return("*****") } elseif (abs(correlation) >=0.5) {return("****") } elseif (abs(correlation) >=0.3) {return("***") } elseif (abs(correlation) >=0.1) {return("**") } else {return("*") }}# panel.cor function with stars alongside correlation coefficientspanel.cor_stars <-function(x, y, digits =2, prefix ="", cex.cor, ...) { usr <-par("usr"); on.exit(par(usr)) par(usr =c(0, 1, 0, 1)) r <-cor(x, y) stars <-evaluateCorrelationStars(r) txt <-paste0(format(c(r, 0.123456789), digits = digits)[1], " ", stars)if(missing(cex.cor)) cex.cor <-0.8/strwidth(txt)text(0.5, 0.5, txt, cex = cex.cor)}pairs(data_question3_2[,c("overallscore", "cases_per_million", "deaths_per_million", "stringency")], upper.panel = panel.cor_stars, diag.panel=panel.hist, lower.panel = lower.panel,main="Correlation table and distribution of COVID variables")
=======
Code
lower.panel <-function(x, y, ...){points(x, y, pch =20, col ="black", cex =0.2)}evaluateCorrelationStars <-function(correlation) {if (abs(correlation) >=0.7) {return("*****") } elseif (abs(correlation) >=0.5) {return("****") } elseif (abs(correlation) >=0.3) {return("***") } elseif (abs(correlation) >=0.1) {return("**") } else {return("*") }}# panel.cor function with stars alongside correlation coefficientspanel.cor_stars <-function(x, y, digits =2, prefix ="", cex.cor, ...) { usr <-par("usr"); on.exit(par(usr)) par(usr =c(0, 1, 0, 1)) r <-cor(x, y) stars <-evaluateCorrelationStars(r) txt <-paste0(format(c(r, 0.123456789), digits = digits)[1], " ", stars)if(missing(cex.cor)) cex.cor <-0.8/strwidth(txt)text(0.5, 0.5, txt, cex = cex.cor)}pairs(data_question3_2[,c("overallscore", "cases_per_million", "deaths_per_million", "stringency")], upper.panel = panel.cor_stars, diag.panel=panel.hist, lower.panel = lower.panel,main="Correlation table and distribution of COVID variables")
>>>>>>> Stashed changes
The different variables used to materialize the impact of COVID19 do not seem to have important influence on the overall score, but we will further explore for the different SDGs, since we believe that COVID19 had a specific influence on some SDGs, for instance “good health and well-being” or “decent work and economic growth”.
<<<<<<< Updated upstream
Code
lower.panel <-function(x, y, ...){points(x, y, pch =20, col ="black", cex =0.5)}evaluateCorrelationStars <-function(correlation) {if (abs(correlation) >=0.7) {return("*****") } elseif (abs(correlation) >=0.5) {return("****") } elseif (abs(correlation) >=0.3) {return("***") } elseif (abs(correlation) >=0.1) {return("**") } else {return("*") }}# panel.cor function with stars alongside correlation coefficientspanel.cor_stars <-function(x, y, digits =2, prefix ="", cex.cor, ...) { usr <-par("usr"); on.exit(par(usr)) par(usr =c(0, 1, 0, 1)) r <-cor(x, y) stars <-evaluateCorrelationStars(r) txt <-paste0(format(c(r, 0.123456789), digits = digits)[1], " ", stars)if(missing(cex.cor)) cex.cor <-0.8/strwidth(txt)text(0.5, 0.5, txt, cex = cex.cor)}pairs(data_question3_3[,c("overallscore", "ongoing", "sum_deaths", "pop_affected", "area_affected", "maxintensity")], upper.panel = panel.cor_stars, diag.panel=panel.hist, lower.panel = lower.panel, main="Correlation table and distribution of conflicts variables")
=======
Code
lower.panel <-function(x, y, ...){points(x, y, pch =20, col ="black", cex =0.5)}evaluateCorrelationStars <-function(correlation) {if (abs(correlation) >=0.7) {return("*****") } elseif (abs(correlation) >=0.5) {return("****") } elseif (abs(correlation) >=0.3) {return("***") } elseif (abs(correlation) >=0.1) {return("**") } else {return("*") }}# panel.cor function with stars alongside correlation coefficientspanel.cor_stars <-function(x, y, digits =2, prefix ="", cex.cor, ...) { usr <-par("usr"); on.exit(par(usr)) par(usr =c(0, 1, 0, 1)) r <-cor(x, y) stars <-evaluateCorrelationStars(r) txt <-paste0(format(c(r, 0.123456789), digits = digits)[1], " ", stars)if(missing(cex.cor)) cex.cor <-0.8/strwidth(txt)text(0.5, 0.5, txt, cex = cex.cor)}pairs(data_question3_3[,c("overallscore", "ongoing", "sum_deaths", "pop_affected", "area_affected", "maxintensity")], upper.panel = panel.cor_stars, diag.panel=panel.hist, lower.panel = lower.panel, main="Correlation table and distribution of conflicts variables")
>>>>>>> Stashed changes
The different variables used to materialize the impact of conflicts do not seem to have important influence on the overall score, but we will further explore for the different SDGs, since we believe that conflicts have a specific influence on some SDGs.
To explore our data on events such as disasters, covid-19 and conflicts we have to first see which countries are the most touched by these. To do so, we made time-series analysis on this three events each time depending on different variables.
<<<<<<< Updated upstream
Code
# Converted 'year' column to date formatQ3.1$year <-as.Date(as.character(Q3.1$year), format ="%Y")Q3.2$year <-as.Date(as.character(Q3.2$year), format ="%Y")Q3.3$year <-as.Date(as.character(Q3.3$year), format ="%Y")
These is our time-analysis concerning the COVID-19 cases per million by region between end 2018 and 2022.
Code
library(ggplot2)covid_filtered <- Q3.2[Q3.2$year >=as.Date("2018-12-12"), ]ggplot(data = covid_filtered, aes(x = year, y = cases_per_million, group = region, color = region)) +geom_smooth(method ="loess", se =FALSE, span =0.8, size =0.5) +labs(title ="Trend of COVID-19 Cases per Million Over Time",x ="Year", y ="Cases per Million") +facet_wrap(~ region, ncol =3) +theme( axis.text.x =element_text(angle =45, size =8, hjust =1),axis.text.y =element_text(vjust =1, size =8, hjust =1),plot.title =element_text(margin =margin(b =20), hjust =0.5, vjust =8, lineheight =2),strip.text =element_blank(),panel.spacing =unit(0.5, "lines") ) +theme(legend.position ="bottom") +guides(color =guide_legend(nrow =3))
=======
Code
# Converted 'year' column to date formatQ3.1$year <-as.Date(as.character(Q3.1$year), format ="%Y")Q3.2$year <-as.Date(as.character(Q3.2$year), format ="%Y")Q3.3$year <-as.Date(as.character(Q3.3$year), format ="%Y")
These is our time-analysis concerning the COVID-19 cases per million by region between end 2018 and 2022.
Code
library(ggplot2)covid_filtered <- Q3.2[Q3.2$year >=as.Date("2018-12-12"), ]ggplot(data = covid_filtered, aes(x = year, y = cases_per_million, group = region, color = region)) +geom_smooth(method ="loess", se =FALSE, span =0.8, size =0.5) +labs(title ="Trend of COVID-19 Cases per Million Over Time",x ="Year", y ="Cases per Million") +facet_wrap(~ region, ncol =3) +theme( axis.text.x =element_text(angle =45, size =8, hjust =1),axis.text.y =element_text(vjust =1, size =8, hjust =1),plot.title =element_text(margin =margin(b =20), hjust =0.5, vjust =8, lineheight =2),strip.text =element_blank(),panel.spacing =unit(0.5, "lines") ) +theme(legend.position ="bottom") +guides(color =guide_legend(nrow =3))
>>>>>>> Stashed changes
These is our time-analysis concerning the COVID-19 deaths per million per region between end 2018 and 2022
<<<<<<< Updated upstream
Code
ggplot(data = covid_filtered, aes(x = year, y = deaths_per_million, group = region, color = region)) +geom_smooth(method ="loess", se =FALSE, span =0.8, size =0.5) +labs(title ="Trend of COVID-19 Deaths per Million Over Time", x ="Year", y ="Deaths per Million") +facet_wrap(~ region, nrow =3) +theme( axis.text.x =element_text(angle =45, size =8, hjust =1),axis.text.y =element_text(vjust =1, size =8, hjust =1),plot.title =element_text(margin =margin(b =20), hjust =0.5, vjust =8, lineheight =2),strip.text =element_blank(),panel.spacing =unit(0.5, "lines") ) +theme(legend.position ="bottom") +guides(color =guide_legend(nlin =3))
=======
Code
ggplot(data = covid_filtered, aes(x = year, y = deaths_per_million, group = region, color = region)) +geom_smooth(method ="loess", se =FALSE, span =0.8, size =0.5) +labs(title ="Trend of COVID-19 Deaths per Million Over Time", x ="Year", y ="Deaths per Million") +facet_wrap(~ region, nrow =3) +theme( axis.text.x =element_text(angle =45, size =8, hjust =1),axis.text.y =element_text(vjust =1, size =8, hjust =1),plot.title =element_text(margin =margin(b =20), hjust =0.5, vjust =8, lineheight =2),strip.text =element_blank(),panel.spacing =unit(0.5, "lines") ) +theme(legend.position ="bottom") +guides(color =guide_legend(nlin =3))
>>>>>>> Stashed changes
These is our time-analysis concerning the COVID-19 stringency per region between end 2018 and 2022
<<<<<<< Updated upstream
Code
ggplot(data = covid_filtered, aes(x = year, y = stringency, group = region, color = region)) +geom_smooth(method ="loess", se =FALSE, span =0.7, size =0.5) +labs(title ="Trend of COVID-19 Stringency Over Time", x ="Year", y ="Stringency") +facet_wrap(~ region, nrow =4) +theme( axis.text.x =element_text(angle =45, size =8, hjust =1),axis.text.y =element_text(vjust =1, size =8, hjust =1),plot.title =element_text(margin =margin(b =20), hjust =0.5, vjust =8, lineheight =2),strip.text =element_blank(),panel.spacing =unit(0.5, "lines") ) +theme(legend.position ="right") +guides(color =guide_legend(ncol =1))
These is our time-analysis concerning climatic disasters with total affected per region
Code
Q3.1[is.na(Q3.1)] <-0ggplot(data = Q3.1, aes(x = year, y = total_affected, group = region, color = region)) +geom_smooth(method ="loess", se =FALSE, span =0.7, size =0.5) +labs(title ="Trend of Total Affected from Climatic Disasters Over Time", x ="Year", y ="Total Affected") +facet_wrap(~ region, nrow =4) +theme( axis.text.x =element_text(angle =45, size =8, hjust =1),axis.text.y =element_text(vjust =1, size =8, hjust =1),plot.title =element_text(margin =margin(b =20), hjust =0.5, vjust =8, lineheight =2),strip.text =element_blank(),panel.spacing =unit(0.5, "lines") ) +theme(legend.position ="right") +guides(color =guide_legend(ncol =1))
=======
Code
ggplot(data = covid_filtered, aes(x = year, y = stringency, group = region, color = region)) +geom_smooth(method ="loess", se =FALSE, span =0.7, size =0.5) +labs(title ="Trend of COVID-19 Stringency Over Time", x ="Year", y ="Stringency") +facet_wrap(~ region, nrow =4) +theme( axis.text.x =element_text(angle =45, size =8, hjust =1),axis.text.y =element_text(vjust =1, size =8, hjust =1),plot.title =element_text(margin =margin(b =20), hjust =0.5, vjust =8, lineheight =2),strip.text =element_blank(),panel.spacing =unit(0.5, "lines") ) +theme(legend.position ="right") +guides(color =guide_legend(ncol =1))
>>>>>>> Stashed changes
<<<<<<< Updated upstream
=======
These is our time-analysis concerning climatic disasters with total affected per region
Code
Q3.1[is.na(Q3.1)] <-0ggplot(data = Q3.1, aes(x = year, y = total_affected, group = region, color = region)) +geom_smooth(method ="loess", se =FALSE, span =0.7, size =0.5) +labs(title ="Trend of Total Affected from Climatic Disasters Over Time", x ="Year", y ="Total Affected") +facet_wrap(~ region, nrow =4) +theme( axis.text.x =element_text(angle =45, size =8, hjust =1),axis.text.y =element_text(vjust =1, size =8, hjust =1),plot.title =element_text(margin =margin(b =20), hjust =0.5, vjust =8, lineheight =2),strip.text =element_blank(),panel.spacing =unit(0.5, "lines") ) +theme(legend.position ="right") +guides(color =guide_legend(ncol =1))
>>>>>>> Stashed changes
These is our time-analysis concerning conflicts deaths per region between 2000 and 2016
<<<<<<< Updated upstream
Code
conflicts_filtered <- Q3.3[Q3.3$year >=as.Date("2000-01-01") & Q3.3$year <=as.Date("2016-12-31"), ]ggplot(data = conflicts_filtered, aes(x = year, y = sum_deaths, group = region, color = region)) +geom_smooth(method ="loess", se =FALSE, span =0.3, size =0.5) +# Using loess smoothing methodlabs(title ="Trend of Deaths by Conflicts Over Time", x ="Year", y ="Sum Deaths") +facet_wrap(~ region, nrow =4) +theme( axis.text.x =element_text(angle =45, size =8, hjust =1),axis.text.y =element_text(vjust =1, size =8, hjust =1),plot.title =element_text(margin =margin(b =20), hjust =0.5, vjust =8, lineheight =2),strip.text =element_blank(),panel.spacing =unit(0.5, "lines") ) +theme(legend.position ="right") +guides(color =guide_legend(ncol =1))
=======
Code
conflicts_filtered <- Q3.3[Q3.3$year >=as.Date("2000-01-01") & Q3.3$year <=as.Date("2016-12-31"), ]ggplot(data = conflicts_filtered, aes(x = year, y = sum_deaths, group = region, color = region)) +geom_smooth(method ="loess", se =FALSE, span =0.3, size =0.5) +# Using loess smoothing methodlabs(title ="Trend of Deaths by Conflicts Over Time", x ="Year", y ="Sum Deaths") +facet_wrap(~ region, nrow =4) +theme( axis.text.x =element_text(angle =45, size =8, hjust =1),axis.text.y =element_text(vjust =1, size =8, hjust =1),plot.title =element_text(margin =margin(b =20), hjust =0.5, vjust =8, lineheight =2),strip.text =element_blank(),panel.spacing =unit(0.5, "lines") ) +theme(legend.position ="right") +guides(color =guide_legend(ncol =1))
>>>>>>> Stashed changes
We can see that the regions’ the most affected by the conflicts are : Middle east and north Africa, Sub-Saharan Africa, South Asia, then less America & the Caribbean and Eastern Europe
These is our time-analysis concerning conflicts affected population per region between 2000 and 2016
<<<<<<< Updated upstream
Code
ggplot(data = conflicts_filtered, aes(x = year, y = pop_affected, group = region, color = region)) +geom_smooth(method ="loess", se =FALSE, span =0.3, size =0.5) +# Using loess smoothing methodlabs(title ="Trend of Population Affected by Conflicts Over Time", x ="Year", y ="pop_affected") +facet_wrap(~ region, nrow =4) +theme( axis.text.x =element_text(angle =45, size =8, hjust =1),axis.text.y =element_text(vjust =1, size =8, hjust =1),plot.title =element_text(margin =margin(b =20), hjust =0.5, vjust =8, lineheight =2),strip.text =element_blank(),panel.spacing =unit(0.5, "lines") ) +theme(legend.position ="right") +guides(color =guide_legend(ncol =1))
=======
Code
ggplot(data = conflicts_filtered, aes(x = year, y = pop_affected, group = region, color = region)) +geom_smooth(method ="loess", se =FALSE, span =0.3, size =0.5) +# Using loess smoothing methodlabs(title ="Trend of Population Affected by Conflicts Over Time", x ="Year", y ="pop_affected") +facet_wrap(~ region, nrow =4) +theme( axis.text.x =element_text(angle =45, size =8, hjust =1),axis.text.y =element_text(vjust =1, size =8, hjust =1),plot.title =element_text(margin =margin(b =20), hjust =0.5, vjust =8, lineheight =2),strip.text =element_blank(),panel.spacing =unit(0.5, "lines") ) +theme(legend.position ="right") +guides(color =guide_legend(ncol =1))
>>>>>>> Stashed changes
We can see that the regions’ the most affected by the conflicts are : Middle east and north Africa, Sub-Saharan Africa, South Asia, America & the Caribbean, Eastern Europe and sometimes Caucasus and Central Asia
Now that we could visualize which regions are the most impacted by these three events we can do correlations analysis per region to see if this events have indeed an impact on the evolution of SDG goals.
Here we want to analyse the correlation between the climate disasters and the SDG goals in South and East Asia.
cor_melted <-as.data.frame(as.table(correlation_matrix_disaster_Asia))names(cor_melted) <-c("Variable1", "Variable2", "Correlation")ggplot(data = cor_melted, aes(Variable1, Variable2, fill = Correlation)) +geom_tile() +scale_fill_gradient2(low ="blue", high ="red", mid ="white",midpoint =0, limit =c(-1, 1), space ="Lab",name ="Correlation") +theme_minimal() +theme(axis.text.x =element_text(angle =45, vjust =1, size =8, hjust =1),axis.text.y =element_text(size =8)) +coord_fixed() +labs(x ='', y ='',title ='Correlation between the climate disasters and the SDG goals in South and East Asia')
=======
cor_melted <-as.data.frame(as.table(correlation_matrix_disaster_Asia))names(cor_melted) <-c("Variable1", "Variable2", "Correlation")ggplot(data = cor_melted, aes(Variable1, Variable2, fill = Correlation)) +geom_tile() +scale_fill_gradient2(low ="blue", high ="red", mid ="white",midpoint =0, limit =c(-1, 1), space ="Lab",name ="Correlation") +theme_minimal() +theme(axis.text.x =element_text(angle =45, vjust =1, size =8, hjust =1),axis.text.y =element_text(size =8)) +coord_fixed() +labs(x ='', y ='',title ='Correlation between the climate disasters and the SDG goals in South and East Asia')
>>>>>>> Stashed changes
We conclude that climate disasters do not really have a big impact on SDG goals.
Here we want to analyse the correlation between the Covid-19 and the SDG goals only during Covid time.
<<<<<<< Updated upstream
Code
covid_filtered <- Q3.2[Q3.2$year >=as.Date("2019-01-01"), ]relevant_columns <-c("goal1", "goal2", "goal3", "goal4", "goal5", "goal6", "goal7", "goal8", "goal9", "goal10", "goal11", "goal12", "goal13", "goal15", "goal16", "stringency", "cases_per_million", "deaths_per_million")# Subset data with relevant columns for correlation analysisrelevant_data <- covid_filtered[, relevant_columns]correlation_matrix_Covid <-cor(relevant_data, use ="complete.obs")kable(correlation_matrix_Covid)
=======
Code
covid_filtered <- Q3.2[Q3.2$year >=as.Date("2019-01-01"), ]relevant_columns <-c("goal1", "goal2", "goal3", "goal4", "goal5", "goal6", "goal7", "goal8", "goal9", "goal10", "goal11", "goal12", "goal13", "goal15", "goal16", "stringency", "cases_per_million", "deaths_per_million")# Subset data with relevant columns for correlation analysisrelevant_data <- covid_filtered[, relevant_columns]correlation_matrix_Covid <-cor(relevant_data, use ="complete.obs")kable(correlation_matrix_Covid)
>>>>>>> Stashed changes
goal1
goal2
goal3
goal4
goal5
goal6
goal7
goal8
goal9
goal10
goal11
goal12
goal13
goal15
goal16
stringency
cases_per_million
deaths_per_million
goal1
1.000
0.534
0.867
0.777
0.445
0.763
0.798
0.584
0.781
0.497
0.727
-0.648
-0.553
0.099
0.714
0.056
0.341
0.361
goal2
0.534
1.000
0.560
0.541
0.469
0.605
0.469
0.636
0.569
0.240
0.463
-0.353
-0.284
0.122
0.451
0.088
0.206
0.242
goal3
0.867
0.560
1.000
0.829
0.641
0.836
0.845
0.693
0.881
0.456
0.828
-0.789
-0.669
0.152
0.825
0.040
0.412
0.373
goal4
0.777
0.541
0.829
1.000
0.656
0.764
0.803
0.596
0.773
0.309
0.758
-0.655
-0.558
0.058
0.674
0.113
0.349
0.339
goal5
0.445
0.469
0.641
0.656
1.000
0.663
0.606
0.587
0.645
0.098
0.690
-0.653
-0.564
0.203
0.628
0.060
0.330
0.261
goal6
0.763
0.605
0.836
0.764
0.663
1.000
0.765
0.711
0.811
0.366
0.766
-0.727
-0.583
0.262
0.729
0.069
0.389
0.398
goal7
0.798
0.469
0.845
0.803
0.606
0.765
1.000
0.556
0.740
0.323
0.793
-0.654
-0.494
0.123
0.697
0.055
0.340
0.374
goal8
0.584
0.636
0.693
0.596
0.587
0.711
0.556
1.000
0.695
0.387
0.587
-0.635
-0.556
0.283
0.627
0.024
0.356
0.278
goal9
0.781
0.569
0.881
0.773
0.645
0.811
0.740
0.695
1.000
0.462
0.753
-0.857
-0.760
0.189
0.819
0.074
0.460
0.353
goal10
0.497
0.240
0.456
0.309
0.098
0.366
0.323
0.387
0.462
1.000
0.281
-0.496
-0.469
0.215
0.519
-0.030
0.262
0.142
goal11
0.727
0.463
0.828
0.758
0.690
0.766
0.793
0.587
0.753
0.281
1.000
-0.696
-0.576
0.089
0.764
0.037
0.345
0.328
goal12
-0.648
-0.353
-0.789
-0.655
-0.653
-0.727
-0.654
-0.635
-0.857
-0.496
-0.696
1.000
0.876
-0.316
-0.825
0.013
-0.466
-0.292
goal13
-0.553
-0.284
-0.669
-0.558
-0.564
-0.583
-0.494
-0.556
-0.760
-0.469
-0.576
0.876
1.000
-0.205
-0.682
-0.018
-0.364
-0.166
goal15
0.099
0.122
0.152
0.058
0.203
0.262
0.123
0.283
0.189
0.215
0.089
-0.316
-0.205
1.000
0.303
-0.068
0.169
0.223
goal16
0.714
0.451
0.825
0.674
0.628
0.729
0.697
0.627
0.819
0.519
0.764
-0.825
-0.682
0.303
1.000
-0.023
0.425
0.316
stringency
0.056
0.088
0.040
0.113
0.060
0.069
0.055
0.024
0.074
-0.030
0.037
0.013
-0.018
-0.068
-0.023
1.000
0.041
0.336
cases_per_million
0.341
0.206
0.412
0.349
0.330
0.389
0.340
0.356
0.460
0.262
0.345
-0.466
-0.364
0.169
0.425
0.041
1.000
0.416
deaths_per_million
0.361
0.242
0.373
0.339
0.261
0.398
0.374
0.278
0.353
0.142
0.328
-0.292
-0.166
0.223
0.316
0.336
0.416
1.000
Code
<<<<<<< Updated upstream
cor_melted <-as.data.frame(as.table(correlation_matrix_Covid))names(cor_melted) <-c("Variable1", "Variable2", "Correlation")ggplot(data = cor_melted, aes(Variable1, Variable2, fill = Correlation)) +geom_tile() +scale_fill_gradient2(low ="blue", high ="red", mid ="white",midpoint =0, limit =c(-1, 1), space ="Lab",name ="Correlation") +theme_minimal() +theme(axis.text.x =element_text(angle =45, vjust =1, size =8, hjust =1),axis.text.y =element_text(size =8)) +coord_fixed() +labs(x ='', y ='',title ='Correlation between COVID and the SDG goals')
=======
cor_melted <-as.data.frame(as.table(correlation_matrix_Covid))names(cor_melted) <-c("Variable1", "Variable2", "Correlation")ggplot(data = cor_melted, aes(Variable1, Variable2, fill = Correlation)) +geom_tile() +scale_fill_gradient2(low ="blue", high ="red", mid ="white",midpoint =0, limit =c(-1, 1), space ="Lab",name ="Correlation") +theme_minimal() +theme(axis.text.x =element_text(angle =45, vjust =1, size =8, hjust =1),axis.text.y =element_text(size =8)) +coord_fixed() +labs(x ='', y ='',title ='Correlation between COVID and the SDG goals')
>>>>>>> Stashed changes
Same conclusion, really weird.
Here we want to analyse the correlation between conflicts deaths and the SDG goals only for the Middle East & North Africa, Sub-Saharan Africa, South Asia, Latin America & the Caribbean and Eastern Europe regions.
<<<<<<< Updated upstream
Code
# Filter data for specific regionsselected_regions <-c("Middle East & North Africa", "Sub-Saharan Africa", "South Asia", "Latin America & the Caribbean", "Eastern Europe")conflicts_selected <- Q3.3[Q3.3$region %in% selected_regions, ]# Select relevant columns for the correlation analysisrelevant_columns <-c("goal1", "goal2", "goal3", "goal4", "goal5", "goal6", "goal7", "goal8", "goal9", "goal10", "goal11", "goal12", "goal13", "goal15", "goal16", "sum_deaths")# Compute correlation matrix for the selected regionscorrelation_matrix_Conflicts_Deaths <-cor(conflicts_selected[, relevant_columns], use ="complete.obs")# View the correlation matrixkable(correlation_matrix_Conflicts_Deaths)
=======
Code
# Filter data for specific regionsselected_regions <-c("Middle East & North Africa", "Sub-Saharan Africa", "South Asia", "Latin America & the Caribbean", "Eastern Europe")conflicts_selected <- Q3.3[Q3.3$region %in% selected_regions, ]# Select relevant columns for the correlation analysisrelevant_columns <-c("goal1", "goal2", "goal3", "goal4", "goal5", "goal6", "goal7", "goal8", "goal9", "goal10", "goal11", "goal12", "goal13", "goal15", "goal16", "sum_deaths")# Compute correlation matrix for the selected regionscorrelation_matrix_Conflicts_Deaths <-cor(conflicts_selected[, relevant_columns], use ="complete.obs")# View the correlation matrixkable(correlation_matrix_Conflicts_Deaths)
>>>>>>> Stashed changes
goal1
goal2
goal3
goal4
goal5
goal6
goal7
goal8
goal9
goal10
goal11
goal12
goal13
goal15
goal16
sum_deaths
goal1
1.000
0.476
0.910
0.791
0.406
0.799
0.865
0.546
0.723
0.272
0.783
-0.730
-0.594
0.039
0.613
-0.095
goal2
0.476
1.000
0.544
0.531
0.540
0.638
0.531
0.571
0.530
0.102
0.475
-0.376
-0.322
0.154
0.430
-0.173
goal3
0.910
0.544
1.000
0.814
0.507
0.832
0.876
0.596
0.768
0.223
0.828
-0.745
-0.587
0.014
0.666
-0.117
goal4
0.791
0.531
0.814
1.000
0.645
0.748
0.808
0.536
0.696
0.089
0.768
-0.667
-0.533
0.007
0.496
-0.101
goal5
0.406
0.540
0.507
0.645
1.000
0.587
0.539
0.454
0.516
-0.178
0.620
-0.464
-0.351
0.191
0.384
-0.162
goal6
0.799
0.638
0.832
0.748
0.587
1.000
0.812
0.670
0.734
0.137
0.788
-0.711
-0.529
0.187
0.599
-0.166
goal7
0.865
0.531
0.876
0.808
0.539
0.812
1.000
0.539
0.720
0.152
0.841
-0.704
-0.531
0.039
0.566
-0.094
goal8
0.546
0.571
0.596
0.536
0.454
0.670
0.539
1.000
0.609
0.209
0.542
-0.519
-0.389
0.181
0.462
-0.102
goal9
0.723
0.530
0.768
0.696
0.516
0.734
0.720
0.609
1.000
0.300
0.698
-0.759
-0.689
0.137
0.591
-0.077
goal10
0.272
0.102
0.223
0.089
-0.178
0.137
0.152
0.209
0.300
1.000
0.035
-0.297
-0.299
0.118
0.275
0.078
goal11
0.783
0.475
0.828
0.768
0.620
0.788
0.841
0.542
0.698
0.035
1.000
-0.729
-0.570
0.031
0.656
-0.155
goal12
-0.730
-0.376
-0.745
-0.667
-0.464
-0.711
-0.704
-0.519
-0.759
-0.297
-0.729
1.000
0.865
-0.170
-0.666
0.122
goal13
-0.594
-0.322
-0.587
-0.533
-0.351
-0.529
-0.531
-0.389
-0.689
-0.299
-0.570
0.865
1.000
-0.150
-0.493
0.079
goal15
0.039
0.154
0.014
0.007
0.191
0.187
0.039
0.181
0.137
0.118
0.031
-0.170
-0.150
1.000
0.191
-0.063
goal16
0.613
0.430
0.666
0.496
0.384
0.599
0.566
0.462
0.591
0.275
0.656
-0.666
-0.493
0.191
1.000
-0.162
sum_deaths
-0.095
-0.173
-0.117
-0.101
-0.162
-0.166
-0.094
-0.102
-0.077
0.078
-0.155
0.122
0.079
-0.063
-0.162
1.000
Code
<<<<<<< Updated upstream
# Melt the correlation matrix for ggplot2cor_melted <-as.data.frame(as.table(correlation_matrix_Conflicts_Deaths))names(cor_melted) <-c("Variable1", "Variable2", "Correlation")# Create the heatmapggplot(data = cor_melted, aes(Variable1, Variable2, fill = Correlation)) +geom_tile() +scale_fill_gradient2(low ="blue", high ="red", mid ="white",midpoint =0, limit =c(-1, 1), space ="Lab",name ="Correlation") +theme_minimal() +theme(axis.text.x =element_text(angle =45, vjust =1, size =8, hjust =1),axis.text.y =element_text(size =8)) +coord_fixed() +labs(x ='', y ='',title ='Correlation between Conflicts deaths and the SDG goals')
=======
# Melt the correlation matrix for ggplot2cor_melted <-as.data.frame(as.table(correlation_matrix_Conflicts_Deaths))names(cor_melted) <-c("Variable1", "Variable2", "Correlation")# Create the heatmapggplot(data = cor_melted, aes(Variable1, Variable2, fill = Correlation)) +geom_tile() +scale_fill_gradient2(low ="blue", high ="red", mid ="white",midpoint =0, limit =c(-1, 1), space ="Lab",name ="Correlation") +theme_minimal() +theme(axis.text.x =element_text(angle =45, vjust =1, size =8, hjust =1),axis.text.y =element_text(size =8)) +coord_fixed() +labs(x ='', y ='',title ='Correlation between Conflicts deaths and the SDG goals')
>>>>>>> Stashed changes
Finally, we want to analyse the correlation between conflicts affected population and the SDG goals only for the Middle East & North Africa, Sub-Saharan Africa, South Asia, Latin America & the Caribbean, Eastern Europe regions and Caucasus and Central Asia.
<<<<<<< Updated upstream
Code
# Filter data for specific regions (pop_affected)selected_regions <-c("Middle East & North Africa", "Sub-Saharan Africa", "South Asia", "Latin America & the Caribbean", "Eastern Europe","Caucasus and Central Asia")conflicts_selected <- Q3.3[Q3.3$region %in% selected_regions, ]# Select relevant columns for the correlation analysisrelevant_columns <-c("goal1", "goal2", "goal3", "goal4", "goal5", "goal6", "goal7", "goal8", "goal9", "goal10", "goal11", "goal12", "goal13", "goal15", "goal16", "pop_affected")# Compute correlation matrix for the selected regionscorrelation_matrix_Conflicts_Pop_Affected <-cor(conflicts_selected[, relevant_columns], use ="complete.obs")# View the correlation matrixkable(correlation_matrix_Conflicts_Pop_Affected)
=======
Code
# Filter data for specific regions (pop_affected)selected_regions <-c("Middle East & North Africa", "Sub-Saharan Africa", "South Asia", "Latin America & the Caribbean", "Eastern Europe","Caucasus and Central Asia")conflicts_selected <- Q3.3[Q3.3$region %in% selected_regions, ]# Select relevant columns for the correlation analysisrelevant_columns <-c("goal1", "goal2", "goal3", "goal4", "goal5", "goal6", "goal7", "goal8", "goal9", "goal10", "goal11", "goal12", "goal13", "goal15", "goal16", "pop_affected")# Compute correlation matrix for the selected regionscorrelation_matrix_Conflicts_Pop_Affected <-cor(conflicts_selected[, relevant_columns], use ="complete.obs")# View the correlation matrixkable(correlation_matrix_Conflicts_Pop_Affected)
>>>>>>> Stashed changes
goal1
goal2
goal3
goal4
goal5
goal6
goal7
goal8
goal9
goal10
goal11
goal12
goal13
goal15
goal16
pop_affected
goal1
1.000
0.476
0.910
0.791
0.406
0.799
0.865
0.546
0.723
0.272
0.783
-0.730
-0.594
0.039
0.613
-0.066
goal2
0.476
1.000
0.544
0.531
0.540
0.638
0.531
0.571
0.530
0.102
0.475
-0.376
-0.322
0.154
0.430
-0.083
goal3
0.910
0.544
1.000
0.814
0.507
0.832
0.876
0.596
0.768
0.223
0.828
-0.745
-0.587
0.014
0.666
-0.058
goal4
0.791
0.531
0.814
1.000
0.645
0.748
0.808
0.536
0.696
0.089
0.768
-0.667
-0.533
0.007
0.496
-0.030
goal5
0.406
0.540
0.507
0.645
1.000
0.587
0.539
0.454
0.516
-0.178
0.620
-0.464
-0.351
0.191
0.384
-0.152
goal6
0.799
0.638
0.832
0.748
0.587
1.000
0.812
0.670
0.734
0.137
0.788
-0.711
-0.529
0.187
0.599
-0.106
goal7
0.865
0.531
0.876
0.808
0.539
0.812
1.000
0.539
0.720
0.152
0.841
-0.704
-0.531
0.039
0.566
-0.071
goal8
0.546
0.571
0.596
0.536
0.454
0.670
0.539
1.000
0.609
0.209
0.542
-0.519
-0.389
0.181
0.462
-0.099
goal9
0.723
0.530
0.768
0.696
0.516
0.734
0.720
0.609
1.000
0.300
0.698
-0.759
-0.689
0.137
0.591
0.000
goal10
0.272
0.102
0.223
0.089
-0.178
0.137
0.152
0.209
0.300
1.000
0.035
-0.297
-0.299
0.118
0.275
0.074
goal11
0.783
0.475
0.828
0.768
0.620
0.788
0.841
0.542
0.698
0.035
1.000
-0.729
-0.570
0.031
0.656
-0.103
goal12
-0.730
-0.376
-0.745
-0.667
-0.464
-0.711
-0.704
-0.519
-0.759
-0.297
-0.729
1.000
0.865
-0.170
-0.666
0.107
goal13
-0.594
-0.322
-0.587
-0.533
-0.351
-0.529
-0.531
-0.389
-0.689
-0.299
-0.570
0.865
1.000
-0.150
-0.493
0.021
goal15
0.039
0.154
0.014
0.007
0.191
0.187
0.039
0.181
0.137
0.118
0.031
-0.170
-0.150
1.000
0.191
-0.108
goal16
0.613
0.430
0.666
0.496
0.384
0.599
0.566
0.462
0.591
0.275
0.656
-0.666
-0.493
0.191
1.000
-0.099
pop_affected
-0.066
-0.083
-0.058
-0.030
-0.152
-0.106
-0.071
-0.099
0.000
0.074
-0.103
0.107
0.021
-0.108
-0.099
1.000
Code
<<<<<<< Updated upstream
# Melt the correlation matrix for ggplot2cor_melted <-as.data.frame(as.table(correlation_matrix_Conflicts_Pop_Affected))names(cor_melted) <-c("Variable1", "Variable2", "Correlation")# Create the heatmapggplot(data = cor_melted, aes(Variable1, Variable2, fill = Correlation)) +geom_tile() +scale_fill_gradient2(low ="blue", high ="red", mid ="white",midpoint =0, limit =c(-1, 1), space ="Lab",name ="Correlation") +theme_minimal() +theme(axis.text.x =element_text(angle =45, vjust =1, size =8, hjust =1),axis.text.y =element_text(size =8)) +coord_fixed() +labs(x ='', y ='',title ='Correlation between Conflicts Affected Population and the SDG goals')
=======
# Melt the correlation matrix for ggplot2cor_melted <-as.data.frame(as.table(correlation_matrix_Conflicts_Pop_Affected))names(cor_melted) <-c("Variable1", "Variable2", "Correlation")# Create the heatmapggplot(data = cor_melted, aes(Variable1, Variable2, fill = Correlation)) +geom_tile() +scale_fill_gradient2(low ="blue", high ="red", mid ="white",midpoint =0, limit =c(-1, 1), space ="Lab",name ="Correlation") +theme_minimal() +theme(axis.text.x =element_text(angle =45, vjust =1, size =8, hjust =1),axis.text.y =element_text(size =8)) +coord_fixed() +labs(x ='', y ='',title ='Correlation between Conflicts Affected Population and the SDG goals')
>>>>>>> Stashed changes
4 Focus on the evolution of SDG scores over time
** How has the adoption of the SDGs in 2015 influenced the achievement of SDGs?
First, we look at the evolution of SDG achievement overall score over time by continent and by region and we see that the general evolution of SDG scores around the world is increasing over the years, but very slowly.
This view that groups the countries by region gives us precision about the previous information. Indeed, it is Western Europe that is particularly above and Sub-Saharan Africa that is clearly below.
Second, we look at the evolution of SDG achievement scores(16) over time for the whole world and by continent. We notice that all SDGs except from goal 9 (industry innovation and infrastructure) are close to one another in terms of level and growth. Goal 9 starts far below the others in 2000 and growths faster until exceeding 50%. In addition, some goals did not increase their scores much in the last two decades, for example goal 13 (climate action) and goal 12 (responsible consumption and production).
<<<<<<< Updated upstream
Code
data4 <- data_question2 %>%group_by(year) %>%summarise(across(starts_with("goal"), mean, na.rm=TRUE)) %>%pivot_longer(cols =starts_with("goal"), names_to ="goal", values_to ="mean_value")color_palette <-c("red", "blue", "green", "orange", "purple", "pink", "lightblue", "gray", "cyan", "magenta", "yellow", "darkgreen", "darkblue", "darkred", "darkgrey", "darkcyan")ggplot(data = data4) +geom_line(mapping =aes(x = year, y = mean_value, color = goal), size =0.7) +geom_point(mapping =aes(x = year, y = mean_value, color = goal), size =1) +scale_color_manual(values = color_palette) +scale_y_continuous(limits =c(0, 100)) +labs(title ="Evolution of the mean SDG achievement scores across the world",y ="Mean SDG Scores",x ="Year" )
=======
Code
data4 <- data_question2 %>%group_by(year) %>%summarise(across(starts_with("goal"), mean, na.rm=TRUE)) %>%pivot_longer(cols =starts_with("goal"), names_to ="goal", values_to ="mean_value")color_palette <-c("red", "blue", "green", "orange", "purple", "pink", "lightblue", "gray", "cyan", "magenta", "yellow", "darkgreen", "darkblue", "darkred", "darkgrey", "darkcyan")ggplot(data = data4) +geom_line(mapping =aes(x = year, y = mean_value, color = goal), size =0.7) +geom_point(mapping =aes(x = year, y = mean_value, color = goal), size =1) +scale_color_manual(values = color_palette) +scale_y_continuous(limits =c(0, 100)) +labs(title ="Evolution of the mean SDG achievement scores across the world",y ="Mean SDG Scores",x ="Year" )
>>>>>>> Stashed changes
We continue with the graph that distinguishes continents to get more information.
<<<<<<< Updated upstream
Code
data5 <- data_question2 %>%group_by(year, continent) %>%summarise(across(starts_with("goal"), mean, na.rm=TRUE)) %>%pivot_longer(cols =starts_with("goal"), names_to ="goal", values_to ="mean_value")ggplot(data = data5) +geom_line(mapping =aes(x = year, y = mean_value, color=continent), size =0.7) +scale_color_manual(values = color_palette) +scale_y_continuous(limits =c(0, 100)) +labs(title ="Evolution of the mean SDG achievement scores by continent",y ="Mean SDG Scores",x ="Years from 2000 to 2022" ) +facet_wrap(~ goal, nrow =4)+scale_x_continuous(breaks =NULL)+theme_light()
=======
Code
data5 <- data_question2 %>%group_by(year, continent) %>%summarise(across(starts_with("goal"), mean, na.rm=TRUE)) %>%pivot_longer(cols =starts_with("goal"), names_to ="goal", values_to ="mean_value")ggplot(data = data5) +geom_line(mapping =aes(x = year, y = mean_value, color=continent), size =0.7) +scale_color_manual(values = color_palette) +scale_y_continuous(limits =c(0, 100)) +labs(title ="Evolution of the mean SDG achievement scores by continent",y ="Mean SDG Scores",x ="Years from 2000 to 2022" ) +facet_wrap(~ goal, nrow =4)+scale_x_continuous(breaks =NULL)+theme_light()
>>>>>>> Stashed changes
We observe that most of the time, Europe is at the top of the graph and Africa at the bottom, except for goals 12 and 13 that are linked to ecology. Some other information stand out:
Americas are far behind the other parts of the world regarding goal 10: reduced inequalities.
Africa is far behind the other continents (even if becoming better) for goals 1, 3, 4 and 7.
Goal 9 (industry, innovation and infrastructure) show exponential growth for almost all continents.
Third we create an interactive map of the world to be able to navigate from year 2000 to 2022, seeing the level of achievement of the SDGs (overall score) for each country.
Again, we see that the overall achievement score of the SDGs is increasing and that the countries that have the most red (bad score) are in Africa. However it is also there that it increases more rapidly. Our hypothesis is that when a score is very low, it is easier to make it better than when it becomes very high (around 90%) it may be hard to increase it, because it would mean perfection. In the next section, we will further investigate this idea.
4.2 Analysis: SDG adoption in 2015
We create one new variable per goal that captures the difference in SDG score between the year of the observation and the previous year. This will allow us to see how the countries improve (or not) on SDG scores each year. In addition, preparing for the specific question around 2015, we only keep the years from 2009 to 2022 (7 years before and after 2015).
Code
<<<<<<< Updated upstream
binary2015 <- data_question2 %>%group_by(code) %>%mutate(across(5:21, ~ . - dplyr::lag(.), .names ="diff_{.col}")) %>%ungroup()# Create a new column (binary variable) with value 1 if the year is after 2015 and zero otherwise. binary2015 <- binary2015 %>%mutate(after2015 =ifelse(year >2015, 1, 0)) %>%filter(as.numeric(year)>=2009)
=======
binary2015 <- data_question2 %>%group_by(code) %>%mutate(across(5:21, ~ . - dplyr::lag(.), .names ="diff_{.col}")) %>%ungroup()# Create a new column (binary variable) with value 1 if the year is after 2015 and zero otherwise. binary2015 <- binary2015 %>%mutate(after2015 =ifelse(year >2015, 1, 0)) %>%filter(as.numeric(year)>=2009)
>>>>>>> Stashed changes
We begin by looking at the distribution of the difference in SDG scores from one year to the next (improvement if it is above zero and deterioration if it is below zero).
Code
<<<<<<< Updated upstream
# histogram of difference in scores between yearsunique_years <-unique(binary2015$year)plot_ly() %>%add_trace(type ="histogram", data = binary2015, x =~diff_overallscore[year ==2009],marker =list(color ="lightgreen", line =list(color ="black", width =1)) ) %>%layout(title ="Distribution of SDG evolution",xaxis =list(title ="Year difference SDG score", range =c(-3, 3)),yaxis =list(title ="Frequency", range =c(0, 40)),sliders =list(list(active =0,currentvalue =list(prefix ="Year: "),steps =lapply(seq_along(unique_years), function(i) { year <- unique_years[i]list(label =as.character(year),method ="restyle",args =list(list(x =list(binary2015$diff_overallscore[binary2015$year == year])) ) ) }) ) ) )
=======
# histogram of difference in scores between yearsunique_years <-unique(binary2015$year)plot_ly() %>%add_trace(type ="histogram", data = binary2015, x =~diff_overallscore[year ==2009],marker =list(color ="lightgreen", line =list(color ="black", width =1)) ) %>%layout(title ="Distribution of SDG evolution",xaxis =list(title ="Year difference SDG score", range =c(-3, 3)),yaxis =list(title ="Frequency", range =c(0, 40)),sliders =list(list(active =0,currentvalue =list(prefix ="Year: "),steps =lapply(seq_along(unique_years), function(i) { year <- unique_years[i]list(label =as.character(year),method ="restyle",args =list(list(x =list(binary2015$diff_overallscore[binary2015$year == year])) ) ) }) ) ) )
>>>>>>> Stashed changes
We notice that across the years, the distribution stays on the right of the x-axis, which means that there are more improvement than deterioration. If there is deterioration, it is less than one percent per year, except some extreme cases, for instance in 2013, there was almost a 3% decrease in the overall SDG score of one country. It is also rare to see improvements of more than 2% per year. Regarding our specific question, we do not see a major improvement of the distribution after 2015, if it was the case we would see the distribution going more to the right, but except for 2017, there are more and more values centered around zero, which means less score improvements overall.
After having visualized the improvements and declines of SDG overall score for the whole world, we are now interested in the top 5 countries in terms of improvement each year and we see that major improvement often comes from Sub-Saharan Africa countries or Middle East and North Africa. This confirms that more efforts are made in these regions to achieve better scores, but we also know from our previous visualizations that their initial scores are lower. Moreover, we record that the higher improvements are of 3% per year and were mostly achieved before 2015. Indeed, we can tell that in terms of maximum improvements, the adoption of SDGs in 2015 did not have a strong impact. We also notice that 2020 is the year with the smallest best improvements. We keep that in mind for the next question regarding events and specifically COVID.
Code
<<<<<<< Updated upstream
top_n_values <-5# Test with ggpot2custom_colors <-c("blue", "darkblue", "cyan", "green", "darkgreen", "lightgreen", "lightblue","turquoise", "lightgrey", "darkgrey")# Get unique regions in the datasetunique_regions <-unique(binary2015$region)# Create a color dictionary mapping each region to a specific colorregion_colors <-setNames(custom_colors[1:length(unique_regions)], unique_regions)library(patchwork)plots <-list()for (year in unique_years) { top_countries <- binary2015[binary2015$year == year, ] %>%arrange(desc(year), desc(diff_overallscore)) %>%head(n = top_n_values) plot <-ggplot(data = top_countries, mapping =aes(x = country, y = diff_overallscore, fill = region)) +geom_bar(stat ="identity") +scale_fill_manual(values = region_colors) +# Use the specified colorslabs(title =paste(year), x =NULL, y =NULL) +theme_minimal() +theme(axis.text.x =element_text(angle =45, vjust =1, hjust =1, size=6), legend.position ="none", plot.title =element_text(size =10)) +scale_y_continuous(limits =c(0, 3)) plots[[as.character(year)]] <- plot}# Arrange the plots in a 4x4 grid using patchworkwrap <-wrap_plots(plots, ncol =5)wrap +plot_annotation(title ='Best 5 countries in terms of SDG score improvement')
=======
top_n_values <-5# Test with ggpot2custom_colors <-c("blue", "darkblue", "cyan", "green", "darkgreen", "lightgreen", "lightblue","turquoise", "lightgrey", "darkgrey")# Get unique regions in the datasetunique_regions <-unique(binary2015$region)# Create a color dictionary mapping each region to a specific colorregion_colors <-setNames(custom_colors[1:length(unique_regions)], unique_regions)library(patchwork)plots <-list()for (year in unique_years) { top_countries <- binary2015[binary2015$year == year, ] %>%arrange(desc(year), desc(diff_overallscore)) %>%head(n = top_n_values) plot <-ggplot(data = top_countries, mapping =aes(x = country, y = diff_overallscore, fill = region)) +geom_bar(stat ="identity") +scale_fill_manual(values = region_colors) +# Use the specified colorslabs(title =paste(year), x =NULL, y =NULL) +theme_minimal() +theme(axis.text.x =element_text(angle =45, vjust =1, hjust =1, size=6), legend.position ="none", plot.title =element_text(size =10)) +scale_y_continuous(limits =c(0, 3)) plots[[as.character(year)]] <- plot}# Arrange the plots in a 4x4 grid using patchworkwrap <-wrap_plots(plots, ncol =5)wrap +plot_annotation(title ='Best 5 countries in terms of SDG score improvement')
>>>>>>> Stashed changes
Code
<<<<<<< Updated upstream
# Create a common legend manuallylegend_data <-data.frame(region = unique_regions)legend_plot <-ggplot(legend_data, aes(x = region, fill = region)) +geom_bar(position =position_stack(reverse =TRUE)) +scale_fill_manual(values = region_colors) +labs(title ="Regions") +theme_void() +theme(legend.position ="none",axis.text.y =element_text(angle =0, hjust =1, size =18),plot.title =element_text(size =20, face ="bold") ) +coord_flip()legend_plot
=======
# Create a common legend manuallylegend_data <-data.frame(region = unique_regions)legend_plot <-ggplot(legend_data, aes(x = region, fill = region)) +geom_bar(position =position_stack(reverse =TRUE)) +scale_fill_manual(values = region_colors) +labs(title ="Regions") +theme_void() +theme(legend.position ="none",axis.text.y =element_text(angle =0, hjust =1, size =18),plot.title =element_text(size =20, face ="bold") ) +coord_flip()legend_plot
>>>>>>> Stashed changes
We continue by looking at the worst 5 countries in terms of decline in SDG overall score each year and we see that the years with the worst declines are those closer to us. Indeed the declines were generally no more than 1%, until 2018, where these became more frequent. We notice that the adoption of SDGs in 2015 may have had a good impact, because during the two years that follow, the worst SDG score declines were low (no more than 1% in 2016 and no more 0.5% in 2017). It was stabilizing, but it was of short duration, because then come the more extreme deteriotations. Interestingly, the regions that had were the worst in terms of decline during the past twelve years were very different, the only pattern appears during the last four years, where most of them are in Latin America and the Caribbean.
Code
<<<<<<< Updated upstream
plots <-list()for (year in unique_years) { top_countries <- binary2015[binary2015$year == year, ] %>%arrange(desc(year), diff_overallscore) %>%head(n = top_n_values) plot <-ggplot(data = top_countries, mapping =aes(x = country, y = diff_overallscore, fill = region)) +geom_bar(stat ="identity") +scale_fill_manual(values = region_colors) +# Use the specified colorslabs(title =paste(year), x =NULL, y =NULL) +theme_minimal() +theme(axis.text.x =element_text(angle =45, vjust =1, hjust =1, size=6), legend.position ="none", plot.title =element_text(size =10)) +scale_y_continuous(limits =c(-3,0)) plots[[as.character(year)]] <- plot}# Arrange the plots in a 4x4 grid using patchworkwrap <-wrap_plots(plots, ncol =5)wrap +plot_annotation(title ='Worst 5 countries in terms of SDG score improvement')
=======
plots <-list()for (year in unique_years) { top_countries <- binary2015[binary2015$year == year, ] %>%arrange(desc(year), diff_overallscore) %>%head(n = top_n_values) plot <-ggplot(data = top_countries, mapping =aes(x = country, y = diff_overallscore, fill = region)) +geom_bar(stat ="identity") +scale_fill_manual(values = region_colors) +# Use the specified colorslabs(title =paste(year), x =NULL, y =NULL) +theme_minimal() +theme(axis.text.x =element_text(angle =45, vjust =1, hjust =1, size=6), legend.position ="none", plot.title =element_text(size =10)) +scale_y_continuous(limits =c(-3,0)) plots[[as.character(year)]] <- plot}# Arrange the plots in a 4x4 grid using patchworkwrap <-wrap_plots(plots, ncol =5)wrap +plot_annotation(title ='Worst 5 countries in terms of SDG score improvement')
>>>>>>> Stashed changes
Code
<<<<<<< Updated upstream
legend_plot
=======
legend_plot
>>>>>>> Stashed changes
We move on to the specific SDG scores and look at the 20 best improvements by score. We additionaly differentiate between the improvements than occurred before and after 2015. We want to see which goals get the best improvements and which countries put more effort into it.
Goals 2 (zero hunger), 3 (good health and well-being), 6 (clean water and sanitation), 8 (decent work and economic growth), 12 (responsible consumption and production), 16 (peace, justice and strong institutions) have very low improvements per year. Indeed, even the best ones are below 10%.
Goal 10 (reduced inequalities) has the best improvements, all 20 best improvements are above 20% and it goes up to 45%.
Some goals clearly had most of their best improvements before 2015: goals 3 (good health and well-being), 5 (gender equality), 6 (clean water and sanitation), 7 (affordable and clean energy).
Some goals clearly had most of their best improvements after 2015: goals 8 (decent work and economic growth), 12 (responsible consumption and production).
Regarding the impact of the adoption of SDGs in 2015, we can not tell that it had a positive impact, because there are not more big improvements after 2015 than before, even a little bit less. In addition, the most impressive improvements mostly occurred before 2015. These conclusions are supported by the next graph: we fit to different regression lines (before and after 2015) to see if there is a jump after the adoption and if the the SDG scores increase faster.
Code
<<<<<<< Updated upstream
# Graphs to show the jump (or not) in 2015# Filter datadata_after_2015 <-filter(binary2015, as.numeric(year) >=2015)data_before_2016 <-filter(binary2015, as.numeric(year) <=2015)plotly::plot_ly() %>% plotly::add_trace(data = data_after_2015, x =~year, y =~fitted(lm(overallscore ~ year, data = data_after_2015)), type ='scatter', mode ='lines', line =list(color ='blue'), name ="After 2015") %>% plotly::add_trace(data = data_before_2016, x =~year, y =~fitted(lm(overallscore ~ year, data = data_before_2016)), type ='scatter', mode ='lines', line =list(color ='red'), name ="Before 2015") %>% plotly::layout(title ="Different patterns across SDGs before and after 2015",xaxis =list(title ="Year"),yaxis =list(title ="SDG achievement score", range =c(30, 85)),shapes =list(list(type ='line',x0 =2015,x1 =2015,y0 =0,y1 =1,yref ='paper',line =list(color ='grey', width =2, dash ='dot') ) ),updatemenus =list(list(buttons =list(list(args =list("y", list(~fitted(lm(overallscore ~ year, data = data_after_2015)),~fitted(lm(overallscore ~ year, data = data_before_2016)) )),label ="Overall score",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal1 ~ year, data = data_after_2015)),~fitted(lm(goal1 ~ year, data = data_before_2016)) )),label ="Goal 1: \nno poverty",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal2 ~ year, data = data_after_2015)),~fitted(lm(goal2 ~ year, data = data_before_2016)) )),label ="Goal 2: \nzero hunger",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal3 ~ year, data = data_after_2015)),~fitted(lm(goal3 ~ year, data = data_before_2016)) )),label ="Goal 3: good health \nand well-being",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal4 ~ year, data = data_after_2015)),~fitted(lm(goal4 ~ year, data = data_before_2016)) )),label ="Goal 4: \nquality education",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal5 ~ year, data = data_after_2015)),~fitted(lm(goal5 ~ year, data = data_before_2016)) )),label ="Goal 5: \ngender equality",method ="restyle" ), list(args =list("y", list(~fitted(lm(goal6 ~ year, data = data_after_2015)),~fitted(lm(goal6 ~ year, data = data_before_2016)) )),label ="Goal 6: clean water \nand sanitation",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal7 ~ year, data = data_after_2015)),~fitted(lm(goal7 ~ year, data = data_before_2016)) )),label ="Goal 7: affordable \nand clean energy",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal8 ~ year, data = data_after_2015)),~fitted(lm(goal8 ~ year, data = data_before_2016)) )),label ="Goal 8: decent work \nand economic growth",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal9 ~ year, data = data_after_2015)),~fitted(lm(goal9 ~ year, data = data_before_2016)) )),label ="Goal 9: industry, innovation \nand infrastructure",method ="restyle" ), list(args =list("y", list(~fitted(lm(goal10 ~ year, data = data_after_2015)),~fitted(lm(goal10 ~ year, data = data_before_2016)) )),label ="Goal 10: \nreduced inequalities",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal11 ~ year, data = data_after_2015)),~fitted(lm(goal11 ~ year, data = data_before_2016)) )),label ="Goal 11: sustainable \ncities and communities",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal12 ~ year, data = data_after_2015)),~fitted(lm(goal12 ~ year, data = data_before_2016)) )),label ="Goal 12: responsible \nconsumption and production",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal13 ~ year, data = data_after_2015)),~fitted(lm(goal13 ~ year, data = data_before_2016)) )),label ="Goal 13: \nclimate action",method ="restyle" ), list(args =list("y", list(~fitted(lm(goal15 ~ year, data = data_after_2015)),~fitted(lm(goal15 ~ year, data = data_before_2016)) )),label ="Goal 15: \nlife on earth",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal16 ~ year, data = data_after_2015)),~fitted(lm(goal16 ~ year, data = data_before_2016)) )),label ="Goal 16: peace, justice \nand strong institutions",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal17 ~ year, data = data_after_2015)),~fitted(lm(goal17 ~ year, data = data_before_2016)) )),label ="Goal 17: partnerships \nfor the goals",method ="restyle" ) ) ) ) )
=======
# Graphs to show the jump (or not) in 2015# Filter datadata_after_2015 <-filter(binary2015, as.numeric(year) >=2015)data_before_2016 <-filter(binary2015, as.numeric(year) <=2015)plotly::plot_ly() %>% plotly::add_trace(data = data_after_2015, x =~year, y =~fitted(lm(overallscore ~ year, data = data_after_2015)), type ='scatter', mode ='lines', line =list(color ='blue'), name ="After 2015") %>% plotly::add_trace(data = data_before_2016, x =~year, y =~fitted(lm(overallscore ~ year, data = data_before_2016)), type ='scatter', mode ='lines', line =list(color ='red'), name ="Before 2015") %>% plotly::layout(title ="Different patterns across SDGs before and after 2015",xaxis =list(title ="Year"),yaxis =list(title ="SDG achievement score", range =c(30, 85)),shapes =list(list(type ='line',x0 =2015,x1 =2015,y0 =0,y1 =1,yref ='paper',line =list(color ='grey', width =2, dash ='dot') ) ),updatemenus =list(list(buttons =list(list(args =list("y", list(~fitted(lm(overallscore ~ year, data = data_after_2015)),~fitted(lm(overallscore ~ year, data = data_before_2016)) )),label ="Overall score",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal1 ~ year, data = data_after_2015)),~fitted(lm(goal1 ~ year, data = data_before_2016)) )),label ="Goal 1: \nno poverty",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal2 ~ year, data = data_after_2015)),~fitted(lm(goal2 ~ year, data = data_before_2016)) )),label ="Goal 2: \nzero hunger",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal3 ~ year, data = data_after_2015)),~fitted(lm(goal3 ~ year, data = data_before_2016)) )),label ="Goal 3: good health \nand well-being",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal4 ~ year, data = data_after_2015)),~fitted(lm(goal4 ~ year, data = data_before_2016)) )),label ="Goal 4: \nquality education",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal5 ~ year, data = data_after_2015)),~fitted(lm(goal5 ~ year, data = data_before_2016)) )),label ="Goal 5: \ngender equality",method ="restyle" ), list(args =list("y", list(~fitted(lm(goal6 ~ year, data = data_after_2015)),~fitted(lm(goal6 ~ year, data = data_before_2016)) )),label ="Goal 6: clean water \nand sanitation",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal7 ~ year, data = data_after_2015)),~fitted(lm(goal7 ~ year, data = data_before_2016)) )),label ="Goal 7: affordable \nand clean energy",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal8 ~ year, data = data_after_2015)),~fitted(lm(goal8 ~ year, data = data_before_2016)) )),label ="Goal 8: decent work \nand economic growth",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal9 ~ year, data = data_after_2015)),~fitted(lm(goal9 ~ year, data = data_before_2016)) )),label ="Goal 9: industry, innovation \nand infrastructure",method ="restyle" ), list(args =list("y", list(~fitted(lm(goal10 ~ year, data = data_after_2015)),~fitted(lm(goal10 ~ year, data = data_before_2016)) )),label ="Goal 10: \nreduced inequalities",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal11 ~ year, data = data_after_2015)),~fitted(lm(goal11 ~ year, data = data_before_2016)) )),label ="Goal 11: sustainable \ncities and communities",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal12 ~ year, data = data_after_2015)),~fitted(lm(goal12 ~ year, data = data_before_2016)) )),label ="Goal 12: responsible \nconsumption and production",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal13 ~ year, data = data_after_2015)),~fitted(lm(goal13 ~ year, data = data_before_2016)) )),label ="Goal 13: \nclimate action",method ="restyle" ), list(args =list("y", list(~fitted(lm(goal15 ~ year, data = data_after_2015)),~fitted(lm(goal15 ~ year, data = data_before_2016)) )),label ="Goal 15: \nlife on earth",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal16 ~ year, data = data_after_2015)),~fitted(lm(goal16 ~ year, data = data_before_2016)) )),label ="Goal 16: peace, justice \nand strong institutions",method ="restyle" ),list(args =list("y", list(~fitted(lm(goal17 ~ year, data = data_after_2015)),~fitted(lm(goal17 ~ year, data = data_before_2016)) )),label ="Goal 17: partnerships \nfor the goals",method ="restyle" ) ) ) ) )
>>>>>>> Stashed changes
Simple OLS regression on the difference between years of SDG scores:
# Create a panel data objectpanel_data <- plm::pdata.frame(binary2015, index =c("country", "year"))# Run the difference-in-differences model to take into account the general evolution over the yearsreg2.2<- plm::plm(diff_overallscore ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.1<- plm::plm(diff_goal1 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.2<- plm::plm(diff_goal2 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.3<- plm::plm(diff_goal3 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.4<- plm::plm(diff_goal4 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.5<- plm::plm(diff_goal5 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.6<- plm::plm(diff_goal6 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.7<- plm::plm(diff_goal7 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.8<- plm::plm(diff_goal8 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.9<- plm::plm(diff_goal9 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.10<- plm::plm(diff_goal10 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.11<- plm::plm(diff_goal11 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.12<- plm::plm(diff_goal12 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.13<- plm::plm(diff_goal13 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.15<- plm::plm(diff_goal15 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.16<- plm::plm(diff_goal16 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.17<- plm::plm(diff_goal17 ~ after2015 + year + after2015:year, data = panel_data,model ="within")# Create a list of your regression modelsmodels_list2 <-list("Overall score"=reg2.2, "Goal 1"=reg2.2.1, "Goal 2"=reg2.2.2, "Goal 3"= reg2.2.3, "Goal 4"=reg2.2.4, "Goal 5"=reg2.2.5, "Goal 6"= reg2.2.6, "Goal 7"=reg2.2.7, "Goal 8"=reg2.2.8, "Goal 9"=reg2.2.9, "Goal 10"=reg2.2.10, "Goal 11"= reg2.2.11,"Goal 12"=reg2.2.12, "Goal 13"=reg2.2.13, "Goal 15"=reg2.2.15, "Goal 16"=reg2.2.16, "Goal 17"=reg2.2.17)huxreg(models_list2[1:9])
=======
# Create a panel data objectpanel_data <- plm::pdata.frame(binary2015, index =c("country", "year"))# Run the difference-in-differences model to take into account the general evolution over the yearsreg2.2<- plm::plm(diff_overallscore ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.1<- plm::plm(diff_goal1 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.2<- plm::plm(diff_goal2 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.3<- plm::plm(diff_goal3 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.4<- plm::plm(diff_goal4 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.5<- plm::plm(diff_goal5 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.6<- plm::plm(diff_goal6 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.7<- plm::plm(diff_goal7 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.8<- plm::plm(diff_goal8 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.9<- plm::plm(diff_goal9 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.10<- plm::plm(diff_goal10 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.11<- plm::plm(diff_goal11 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.12<- plm::plm(diff_goal12 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.13<- plm::plm(diff_goal13 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.15<- plm::plm(diff_goal15 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.16<- plm::plm(diff_goal16 ~ after2015 + year + after2015:year, data = panel_data,model ="within")reg2.2.17<- plm::plm(diff_goal17 ~ after2015 + year + after2015:year, data = panel_data,model ="within")# Create a list of your regression modelsmodels_list2 <-list("Overall score"=reg2.2, "Goal 1"=reg2.2.1, "Goal 2"=reg2.2.2, "Goal 3"= reg2.2.3, "Goal 4"=reg2.2.4, "Goal 5"=reg2.2.5, "Goal 6"= reg2.2.6, "Goal 7"=reg2.2.7, "Goal 8"=reg2.2.8, "Goal 9"=reg2.2.9, "Goal 10"=reg2.2.10, "Goal 11"= reg2.2.11,"Goal 12"=reg2.2.12, "Goal 13"=reg2.2.13, "Goal 15"=reg2.2.15, "Goal 16"=reg2.2.16, "Goal 17"=reg2.2.17)huxreg(models_list2[1:9])
>>>>>>> Stashed changes
Overall score
Goal 1
Goal 2
Goal 3
Goal 4
Goal 5
Goal 6
Goal 7
Goal 8
after2015
-0.298 ***
1.036 ***
-0.449 *
-0.730 ***
-0.390
-0.382
-0.306 ***
-0.645 ***
0.788 ***
(0.062)
(0.218)
(0.194)
(0.119)
(0.328)
(0.203)
(0.041)
(0.179)
(0.113)
year2010
0.021
0.000
-0.523 **
1.019 ***
0.359
0.023
-0.057
-0.164
0.303 **
(0.062)
(0.218)
(0.194)
(0.119)
(0.328)
(0.203)
(0.041)
(0.179)
(0.113)
year2011
-0.045
0.878 ***
-0.089
-0.108
0.424
-0.056
-0.023
-0.355 *
0.446 ***
(0.062)
(0.218)
(0.194)
(0.119)
(0.328)
(0.203)
(0.041)
(0.179)
(0.113)
year2012
0.087
0.762 ***
-0.543 **
-0.179
0.379
0.288
-0.062
0.017
0.450 ***
(0.062)
(0.218)
(0.194)
(0.119)
(0.328)
(0.203)
(0.041)
(0.179)
(0.113)
year2013
-0.005
1.015 ***
-0.190
-0.309 **
-0.045
1.259 ***
-0.020
-0.057
0.417 ***
(0.062)
(0.218)
(0.194)
(0.119)
(0.328)
(0.203)
(0.041)
(0.179)
(0.113)
year2014
0.157 *
0.508 *
0.086
-0.295 *
0.730 *
0.101
-0.052
-0.447 *
1.452 ***
(0.062)
(0.218)
(0.194)
(0.119)
(0.328)
(0.203)
(0.041)
(0.179)
(0.113)
year2015
0.001
0.580 **
-0.735 ***
0.325 **
0.038
0.213
-0.002
0.061
0.536 ***
(0.062)
(0.218)
(0.194)
(0.119)
(0.328)
(0.203)
(0.041)
(0.179)
(0.113)
year2016
0.207 ***
-0.710 **
-0.128
0.432 ***
0.447
0.314
0.233 ***
0.370 *
-0.252 *
(0.062)
(0.218)
(0.194)
(0.119)
(0.328)
(0.203)
(0.041)
(0.179)
(0.113)
year2017
0.589 ***
-0.558 *
0.021
0.900 ***
0.572
0.497 *
0.263 ***
0.470 **
0.821 ***
(0.062)
(0.218)
(0.194)
(0.119)
(0.328)
(0.203)
(0.041)
(0.179)
(0.113)
year2018
0.254 ***
-0.594 **
0.210
0.501 ***
0.799 *
0.264
0.208 ***
0.351 *
-0.187
(0.062)
(0.218)
(0.194)
(0.119)
(0.328)
(0.203)
(0.041)
(0.179)
(0.113)
year2019
0.337 ***
-0.831 ***
0.295
0.868 ***
0.474
0.614 **
0.159 ***
0.270
-0.262 *
(0.062)
(0.218)
(0.194)
(0.119)
(0.328)
(0.203)
(0.041)
(0.179)
(0.113)
year2020
0.127 *
-1.874 ***
-0.045
-0.135
0.388
0.167
0.179 ***
0.296
-1.159 ***
(0.062)
(0.218)
(0.194)
(0.119)
(0.328)
(0.203)
(0.041)
(0.179)
(0.113)
year2021
0.274 ***
-0.294
-0.015
0.032
0.010
0.441 *
-0.000
0.000
0.545 ***
(0.062)
(0.218)
(0.194)
(0.119)
(0.328)
(0.203)
(0.041)
(0.179)
(0.113)
N
2170
2002
2170
2170
2170
2170
2170
2170
2170
R2
0.060
0.068
0.020
0.175
0.011
0.043
0.070
0.020
0.228
*** p < 0.001; ** p < 0.01; * p < 0.05.
Code
<<<<<<< Updated upstream
huxreg(models_list2[10:17])
=======
huxreg(models_list2[10:17])
>>>>>>> Stashed changes
Goal 9
Goal 10
Goal 11
Goal 12
Goal 13
Goal 15
Goal 16
Goal 17
after2015
-0.654 **
-1.080 *
-0.394
-0.292 ***
-0.395 **
-0.405 *
-1.150 ***
0.032
(0.231)
(0.500)
(0.309)
(0.087)
(0.151)
(0.163)
(0.173)
(0.229)
year2010
0.174
-0.483
0.127
-0.319 ***
-0.668 ***
0.246
0.617 ***
-0.395
(0.231)
(0.500)
(0.309)
(0.087)
(0.151)
(0.163)
(0.173)
(0.229)
year2011
0.460 *
-0.798
-0.783 *
-0.422 ***
-0.688 ***
-0.114
0.486 **
-0.234
(0.231)
(0.500)
(0.309)
(0.087)
(0.151)
(0.163)
(0.173)
(0.229)
year2012
1.231 ***
-1.010 *
0.218
-0.178 *
-0.127
-0.188
0.230
-0.105
(0.231)
(0.500)
(0.309)
(0.087)
(0.151)
(0.163)
(0.173)
(0.229)
year2013
0.804 ***
-0.764
-0.474
-0.369 ***
-0.366 *
-0.060
-0.475 **
-0.255
(0.231)
(0.500)
(0.309)
(0.087)
(0.151)
(0.163)
(0.173)
(0.229)
year2014
0.933 ***
-0.794
0.618 *
-0.254 **
-0.244
-0.185
0.852 ***
-0.205
(0.231)
(0.500)
(0.309)
(0.087)
(0.151)
(0.163)
(0.173)
(0.229)
year2015
0.725 **
-0.276
-0.619 *
-0.128
-0.014
-0.250
0.493 **
-0.837 ***
(0.231)
(0.500)
(0.309)
(0.087)
(0.151)
(0.163)
(0.173)
(0.229)
year2016
0.596 *
0.155
1.254 ***
-0.042
0.299 *
0.203
0.964 ***
-0.561 *
(0.231)
(0.500)
(0.309)
(0.087)
(0.151)
(0.163)
(0.173)
(0.229)
year2017
3.605 ***
1.026 *
0.409
-0.070
-0.021
0.285
1.529 ***
0.479 *
(0.231)
(0.500)
(0.309)
(0.087)
(0.151)
(0.163)
(0.173)
(0.229)
year2018
1.037 ***
0.443
0.128
-0.040
-0.169
-0.743 ***
1.901 ***
0.494 *
(0.231)
(0.500)
(0.309)
(0.087)
(0.151)
(0.163)
(0.173)
(0.229)
year2019
1.208 ***
0.631
0.063
-0.072
0.164
0.201
1.433 ***
-0.035
(0.231)
(0.500)
(0.309)
(0.087)
(0.151)
(0.163)
(0.173)
(0.229)
year2020
1.235 ***
0.283
-0.135
0.009
0.452 **
0.096
1.503 ***
1.153 ***
(0.231)
(0.500)
(0.309)
(0.087)
(0.151)
(0.163)
(0.173)
(0.229)
year2021
1.148 ***
-0.008
-0.181
0.568 ***
-0.164
0.671 ***
1.099 ***
1.179 ***
(0.231)
(0.500)
(0.309)
(0.087)
(0.151)
(0.163)
(0.173)
(0.229)
N
2170
2002
2170
2170
2170
2170
2170
2170
R2
0.140
0.007
0.031
0.056
0.035
0.052
0.109
0.082
*** p < 0.001; ** p < 0.01; * p < 0.05.
5 Analysis
5.1 Answers to the research questions
5.1.1 Focus on relationship between SDGs
How are the different SDGs linked? (We want to see if some SDGs are linked in the fact that a high score on one implies a high score on the other, and thus if we can make groups of SDGs that are comparable in that way).
5.2 Focus on relationship between SDGs
Let’s analyse our relationship between the SDGs. For that, we’ll import our dataset to examine the interconnections among the Sustainable Development Goals (SDGs). After importing, we’ll focus specifically on the columns representing the goals of interest. To provide a comprehensive analysis, we will construct a correlation matrix, highlighting only those goals where the correlation coefficient is either greater than 0.5 (indicating a strong positive relationship) or less than -0.5 (signifying a strong negative relationship). This approach will enable us to identify and analyze the most significant relationships between the selected SDGs.
Given that our variables do not follow a normal distribution, employing the Pearson correlation method is not suitable in our analysis. We attempted to normalize the data through logarithmic and square root transformations, but these adjustments were insufficiently effective. Consequently, we will resort to computing the Spearman correlation. While not ideal, this method does not necessitate the normal distribution of our data. In our analysis, particularly for the heatmap visualization, we will focus on correlations that exceed the threshold of r threshold_heatmap or fall below -r threshold_heatmap. This selective approach will enhance the readability and interpretability of the heatmap.
Code
<<<<<<< Updated upstream
spearman_corr_4_cl <-cor(goals_data_4_cl, method ="spearman", use ="everything")spearman_corr_4_cl[abs(spearman_corr_4_cl) < threashold_heatmap] <-NA
=======
spearman_corr_4_cl <-cor(goals_data_4_cl, method ="spearman", use ="everything")spearman_corr_4_cl[abs(spearman_corr_4_cl) < threashold_heatmap] <-NA
>>>>>>> Stashed changes
We can then plot the Heatmap of the Spearman correlation using the ggplot2 package.
Code
<<<<<<< Updated upstream
# Melting the datamelted_corr_4 <-melt(spearman_corr_4_cl, na.rm =TRUE)# Creating the heatmapggplot(data = melted_corr_4, aes(x = Var1, y = Var2, fill = value)) +geom_tile() +geom_text(aes(label =sprintf("%.2f", value)), vjust =0.5, size=2.5) +# Adding textscale_fill_gradient2(low ="blue", high ="red", mid ="white", midpoint =0, limit =c(-1,1), space ="Lab", name="Spearman\nCorrelation",na.value ="grey") +theme_minimal() +theme(axis.text.x =element_text(angle =45, hjust =1)) +labs(title ="Heatmap of Spearman Correlations for Goals", x ="", y ="")
=======
# Melting the datamelted_corr_4 <-melt(spearman_corr_4_cl, na.rm =TRUE)# Creating the heatmapggplot(data = melted_corr_4, aes(x = Var1, y = Var2, fill = value)) +geom_tile() +geom_text(aes(label =sprintf("%.2f", value)), vjust =0.5, size=2.5) +# Adding textscale_fill_gradient2(low ="blue", high ="red", mid ="white", midpoint =0, limit =c(-1,1), space ="Lab", name="Spearman\nCorrelation",na.value ="grey") +theme_minimal() +theme(axis.text.x =element_text(angle =45, hjust =1)) +labs(title ="Heatmap of Spearman Correlations for Goals", x ="", y ="")
>>>>>>> Stashed changes
It is evident that the Sustainable Development Goals (SDGs) are intricately interconnected. However, certain goals appear to be less interrelated compared to others. Specifically, SDG 1 (No Poverty) and SDG 10 (Reduced Inequalities) demonstrate a weaker correlation with the rest of the goals. Similarly, Goal 15 (Life on Land) also exhibits a lesser degree of interconnection with the other SDGs.
Code
<<<<<<< Updated upstream
# Selecting only numeric columns, assuming they are named as 'goal1', 'goal2', etc.goals_data <- goals_data_4_cl[, grep('goal', names(goals_data_4_cl))]goals_data_scaled <-scale(goals_data) # Scaling the datapca_result <-prcomp(goals_data_scaled) # Running PCA# Plotting Scree plot to visualize the importance of each principal componentfviz_eig(pca_result,addlabels =TRUE) +theme_minimal()# Plotting Biplot to visualize the two main PCsfviz_pca_biplot(pca_result,label="var",col.var="dodgerblue3",geom="point",pointsize =0.1,labelsize =5) +theme_minimal()
=======
# Selecting only numeric columns, assuming they are named as 'goal1', 'goal2', etc.goals_data <- goals_data_4_cl[, grep('goal', names(goals_data_4_cl))]goals_data_scaled <-scale(goals_data) # Scaling the datapca_result <-prcomp(goals_data_scaled) # Running PCA# Summary of PCA - shows variance explained by each principal component# summary(pca_result)# Plotting Scree plot to visualize the importance of each principal componentfviz_eig(pca_result,addlabels =TRUE) +theme_minimal()# Plotting Biplot to visualize the two main PCsfviz_pca_biplot(pca_result,label="var",col.var="dodgerblue3",geom="point",pointsize =0.1,labelsize =5) +theme_minimal()
>>>>>>> Stashed changes
In our EDA on the focus on the influence of the factors over the SDG scores, we had made a correlation matrix heatmap that took into account every variable of our dataset. Here, we tried to zoom on certain parts of the heatmap. We have decided to add on our graphs the correlations between variables when our pvalue was significant (alpha = 0.05). The grey zones are concerning our non-significant pvalues.
Let’s see first the correlation matrix heatmap regarding our SDG goals and all our variables different than our SDG goals.
<<<<<<< Updated upstream
Code
corr_matrix <-cor(data_question1[7:40])p_matrix2 <-matrix(nrow =ncol(data_question1[7:40]), ncol =ncol(data_question1[7:40]))for (i in1:ncol(data_question1[7:40])) {for (j in1:ncol(data_question1[7:40])) { test_result <-cor.test(data_question1[7:40][, i], data_question1[7:40][, j]) p_matrix2[i, j] <- test_result$p.value }}#Switch population at the end of heatmapcorr_matrix[which(p_matrix2 >0.05)] <-NAmelted_corr_matrix_GVar <-melt(corr_matrix[19:34,1:18])ggplot(melted_corr_matrix_GVar, aes(Var1, Var2, fill = value)) +geom_tile() +geom_text(aes(label =ifelse(!is.na(value), sprintf("%.2f", value), '')),color ="black", size =2) +scale_fill_gradient2(low ="blue", high ="red", mid ="white",midpoint =0, limit =c(-1, 1), space ="Lab",name ="Pearson\nCorrelation") +theme_minimal() +theme(axis.text.x =element_text(angle =45, hjust =1),axis.text.y =element_text(angle =45, hjust =1)) +labs(x ='Goals', y ='Goals',title ='Correlations Heatmap between goals and our other variables')
=======
Code
corr_matrix <-cor(data_question1[7:40])p_matrix2 <-matrix(nrow =ncol(data_question1[7:40]), ncol =ncol(data_question1[7:40]))for (i in1:ncol(data_question1[7:40])) {for (j in1:ncol(data_question1[7:40])) { test_result <-cor.test(data_question1[7:40][, i], data_question1[7:40][, j]) p_matrix2[i, j] <- test_result$p.value }}#Switch population at the end of heatmapcorr_matrix[which(p_matrix2 >0.05)] <-NAmelted_corr_matrix_GVar <-melt(corr_matrix[19:34,1:18])ggplot(melted_corr_matrix_GVar, aes(Var1, Var2, fill = value)) +geom_tile() +geom_text(aes(label =ifelse(!is.na(value), sprintf("%.2f", value), '')),color ="black", size =2) +scale_fill_gradient2(low ="blue", high ="red", mid ="white",midpoint =0, limit =c(-1, 1), space ="Lab",name ="Pearson\nCorrelation") +theme_minimal() +theme(axis.text.x =element_text(angle =45, hjust =1),axis.text.y =element_text(angle =45, hjust =1)) +labs(x ='Goals', y ='Goals',title ='Correlations Heatmap between goals and our other variables')
>>>>>>> Stashed changes
As we can see, our SDG goals 12 & 13 (responsible consumption & production and climate action) are negatively correlated with most of our variables, as is the economic freedom government variable to our SDG goals. In that sens, we could understand it as having a higher Human Freedom Index Score would influence more negatively the SDG scores of these two goals, i.e. the more people in a country can access and afford civil justice, the more it impact negatively the score of these two SDG goals.
Nevertheless, goals 12 & 13 and ef_government are positively correlated together. In addition, some variables such as internet_usage, pf_law or ef_legal are strongely correlated with most of our SDG goals. This is mostly due to the large scope englobed in these variables. That makes them influence various sectors of our economies and thus, mostly impacting all our SDG goals.
<<<<<<< Updated upstream
Now let’s zoom on the correlations between all our variables except our SDG goals: ::: {.cell layout-align=“center” hash=‘report_cache/html/unnamed-chunk-299_bdc59270cadbaf59cd1e582101818664’}
Code
melted_corr_matrix_Var <-melt(corr_matrix[19:34,19:34])ggplot(melted_corr_matrix_Var, aes(Var1, Var2, fill = value)) +geom_tile() +geom_text(aes(label =ifelse(!is.na(value), sprintf("%.2f", value), '')),color ="black", size =1.7) +scale_fill_gradient2(low ="blue", high ="red", mid ="white",midpoint =0, limit =c(-1, 1), space ="Lab",name ="Pearson\nCorrelation") +theme_minimal() +theme(axis.text.x =element_text(angle =45, hjust =1),axis.text.y =element_text(angle =45, hjust =1)) +labs(x ='Goals', y ='Goals',title ='Correlations Heatmap between other variables than SDG goals')
=======
Now let’s zoom on the correlations between all our variables except our SDG goals: ::: {.cell layout-align=“center” hash=‘report_cache/html/unnamed-chunk-296_7972c6db1674c30dcabe7d6f14ba43fc’}
Code
melted_corr_matrix_Var <-melt(corr_matrix[19:34,19:34])ggplot(melted_corr_matrix_Var, aes(Var1, Var2, fill = value)) +geom_tile() +geom_text(aes(label =ifelse(!is.na(value), sprintf("%.2f", value), '')),color ="black", size =1.7) +scale_fill_gradient2(low ="blue", high ="red", mid ="white",midpoint =0, limit =c(-1, 1), space ="Lab",name ="Pearson\nCorrelation") +theme_minimal() +theme(axis.text.x =element_text(angle =45, hjust =1),axis.text.y =element_text(angle =45, hjust =1)) +labs(x ='Goals', y ='Goals',title ='Correlations Heatmap between other variables than SDG goals')
Concerning the SDG goals, we conclude that most of our variables are going along the 1st component, except the goals 10 and 15 that are rather uncorrelated with the dimension 1. In addition, as seen before, the goals 12 and 13 are negatively correlated to the other goals. With a eigenvalue bigger than 1 for the first two components, we conclude that there are only 2 dimensions to take into account, according to the Kaiser-Guttman’s rule. Nevertheless, they are explaining less than 80% of cumulated variance.
Now concerning the Human Freedom Index scores, most of the variables are positively correlated to the dimension 1, slightly less for the PF religion and security, and finaly the EF government variable is uncorrelated to the dimension 1. With a eigenvalue bigger than 1 for the three first components, we conclude that there are 3 dimensions to take into account. Nevertheless, again, they are explaining less than 80% of cumulated variance.
Due to the large number of data, the visualization of the clusters using the kmean method is not really relevant. In addition, by clustering our data, we are trying to get group that differ from eachother but with little variation of the observations within the same cluster. Here, only 60.6% of the variance is explained by the variation between clusters. This is not enough.
<<<<<<< Updated upstream
We have noticed that we had high multicolinearity in on regression. Therefore, before to compute them, let’s try to get rid of on of the two variables having at least |0.8| of correlation. ::: {.cell layout-align=“center” hash=‘report_cache/html/unnamed-chunk-303_260f9d8ccafe0462ee5770ac59a91daa’}
Code
correlation_overall_matrix <-cor(Correlation_overall, use ="everything")high_cor_pairs <-which(abs(correlation_overall_matrix) >=0.8, arr.ind =TRUE)# Displaying the resultsfor (i in1:nrow(high_cor_pairs)) { row <- high_cor_pairs[i, "row"] col <- high_cor_pairs[i, "col"]# Avoiding duplicate pairs and diagonal elementsif (row < col) {cat(sprintf("Variables: %s and %s, Correlation: %f\n", names(Correlation_overall)[row], names(Correlation_overall)[col], correlation_overall_matrix[row, col])) }}#> Variables: overallscore and goal1, Correlation: 0.890397#> Variables: overallscore and goal3, Correlation: 0.942610#> Variables: goal1 and goal3, Correlation: 0.894344#> Variables: overallscore and goal4, Correlation: 0.872865#> Variables: goal1 and goal4, Correlation: 0.822114#> Variables: goal3 and goal4, Correlation: 0.851160#> Variables: overallscore and goal6, Correlation: 0.904641#> Variables: goal3 and goal6, Correlation: 0.874924#> Variables: overallscore and goal7, Correlation: 0.901153#> Variables: goal1 and goal7, Correlation: 0.867157#> Variables: goal3 and goal7, Correlation: 0.879306#> Variables: goal4 and goal7, Correlation: 0.822145#> Variables: goal6 and goal7, Correlation: 0.829524#> Variables: overallscore and goal9, Correlation: 0.834913#> Variables: goal3 and goal9, Correlation: 0.811679#> Variables: overallscore and goal11, Correlation: 0.885962#> Variables: goal3 and goal11, Correlation: 0.873018#> Variables: goal4 and goal11, Correlation: 0.831999#> Variables: goal6 and goal11, Correlation: 0.823253#> Variables: goal7 and goal11, Correlation: 0.841304#> Variables: goal9 and goal12, Correlation: -0.836629#> Variables: goal12 and goal13, Correlation: 0.887101#> Variables: overallscore and goal16, Correlation: 0.814057#> Variables: goal12 and goal16, Correlation: -0.818053#> Variables: goal9 and GDPpercapita, Correlation: 0.811971#> Variables: goal12 and GDPpercapita, Correlation: -0.848970#> Variables: overallscore and internet_usage, Correlation: 0.804896#> Variables: goal9 and internet_usage, Correlation: 0.891350#> Variables: goal12 and pf_law, Correlation: -0.850543#> Variables: goal16 and pf_law, Correlation: 0.841794#> Variables: pf_religion and pf_assembly, Correlation: 0.845992#> Variables: pf_assembly and pf_expression, Correlation: 0.888191#> Variables: goal9 and ef_legal, Correlation: 0.829291#> Variables: goal12 and ef_legal, Correlation: -0.837627#> Variables: goal16 and ef_legal, Correlation: 0.839944#> Variables: pf_law and ef_legal, Correlation: 0.852443# List of high-correlation pairscorrelation_pairs <-list(c("overallscore", "goal1"), c("overallscore", "goal3"), c("goal1", "goal3"),c("overallscore", "goal4"), c("goal1", "goal4"), c("goal3", "goal4"),c("overallscore", "goal6"), c("goal3", "goal6"),c("overallscore", "goal7"), c("goal1", "goal7"), c("goal3", "goal7"), c("goal4", "goal7"), c("goal6", "goal7"),c("overallscore", "goal9"), c("goal3", "goal9"),c("overallscore", "goal11"), c("goal3", "goal11"), c("goal4", "goal11"), c("goal6", "goal11"), c("goal7", "goal11"),c("goal9", "goal12"), c("goal12", "goal13"),c("overallscore", "goal16"), c("goal12", "goal16"),c("goal9", "GDPpercapita"), c("goal12", "GDPpercapita"),c("overallscore", "internet_usage"), c("goal9", "internet_usage"),c("goal12", "pf_law"), c("goal16", "pf_law"),c("pf_religion", "pf_assembly"), c("pf_assembly", "pf_expression"),c("goal9", "ef_legal"), c("goal12", "ef_legal"), c("goal16", "ef_legal"), c("pf_law", "ef_legal"))# Flatten the list and count the frequency of each variableflattened_list <-unlist(correlation_pairs)frequency_count <-table(flattened_list)variables_to_remove <-c()for (pair in correlation_pairs) {# Select the variable that appears more frequently for removalif (frequency_count[pair[1]] > frequency_count[pair[2]]) { variables_to_remove <-c(variables_to_remove, pair[1]) } elseif (frequency_count[pair[1]] < frequency_count[pair[2]]) { variables_to_remove <-c(variables_to_remove, pair[2]) } else {# If both appear equally, arbitrarily choose one to remove variables_to_remove <-c(variables_to_remove, pair[1]) }}variables_to_remove <-unique(variables_to_remove)variables_to_remove <-sort(variables_to_remove)print(variables_to_remove) #> [1] "ef_legal" "goal11" "goal12" "goal16" #> [5] "goal3" "goal4" "goal7" "goal9" #> [9] "overallscore" "pf_assembly"
::: Therefore, we will not take into account the variables “ef_legal” “goal11” “goal12” “goal16” “goal3” “goal4” “goal7” “goal9” “overallscore” “pf_assembly” in our regressions, for multicollinearity purpose.
Now, let’s compute the regressions without these variables. ::: {.cell layout-align=“center” hash=‘report_cache/html/unnamed-chunk-304_b8c76702ff167b60ed6666b7ba5a6d3e’}
::: The problem is that even by getting rid of the previous variables, there still might be multicollinearity. Therefore, we need to analyse the vif for each regression and adapt the model in consequence. ::: {.cell layout-align=“center” hash=‘report_cache/html/unnamed-chunk-305_3077c7af945c7583aee4a61f46177929’}
We have noticed that we had high multicolinearity in on regression. Therefore, before to compute them, let’s try to get rid of on of the two variables having at least |0.8| of correlation. ::: {.cell layout-align=“center” hash=‘report_cache/html/unnamed-chunk-300_1d971ce600370dc875ba0dfefc2d6276’}
Code
correlation_overall_matrix <-cor(Correlation_overall, use ="everything")high_cor_pairs <-which(abs(correlation_overall_matrix) >=0.8, arr.ind =TRUE)# Displaying the resultsfor (i in1:nrow(high_cor_pairs)) { row <- high_cor_pairs[i, "row"] col <- high_cor_pairs[i, "col"]# Avoiding duplicate pairs and diagonal elementsif (row < col) {cat(sprintf("Variables: %s and %s, Correlation: %f\n", names(Correlation_overall)[row], names(Correlation_overall)[col], correlation_overall_matrix[row, col])) }}#> Variables: overallscore and goal1, Correlation: 0.890397#> Variables: overallscore and goal3, Correlation: 0.942610#> Variables: goal1 and goal3, Correlation: 0.894344#> Variables: overallscore and goal4, Correlation: 0.872865#> Variables: goal1 and goal4, Correlation: 0.822114#> Variables: goal3 and goal4, Correlation: 0.851160#> Variables: overallscore and goal6, Correlation: 0.904641#> Variables: goal3 and goal6, Correlation: 0.874924#> Variables: overallscore and goal7, Correlation: 0.901153#> Variables: goal1 and goal7, Correlation: 0.867157#> Variables: goal3 and goal7, Correlation: 0.879306#> Variables: goal4 and goal7, Correlation: 0.822145#> Variables: goal6 and goal7, Correlation: 0.829524#> Variables: overallscore and goal9, Correlation: 0.834913#> Variables: goal3 and goal9, Correlation: 0.811679#> Variables: overallscore and goal11, Correlation: 0.885962#> Variables: goal3 and goal11, Correlation: 0.873018#> Variables: goal4 and goal11, Correlation: 0.831999#> Variables: goal6 and goal11, Correlation: 0.823253#> Variables: goal7 and goal11, Correlation: 0.841304#> Variables: goal9 and goal12, Correlation: -0.836629#> Variables: goal12 and goal13, Correlation: 0.887101#> Variables: overallscore and goal16, Correlation: 0.814057#> Variables: goal12 and goal16, Correlation: -0.818053#> Variables: goal9 and GDPpercapita, Correlation: 0.811971#> Variables: goal12 and GDPpercapita, Correlation: -0.848970#> Variables: overallscore and internet_usage, Correlation: 0.804896#> Variables: goal9 and internet_usage, Correlation: 0.891350#> Variables: goal12 and pf_law, Correlation: -0.850543#> Variables: goal16 and pf_law, Correlation: 0.841794#> Variables: pf_religion and pf_assembly, Correlation: 0.845992#> Variables: pf_assembly and pf_expression, Correlation: 0.888191#> Variables: goal9 and ef_legal, Correlation: 0.829291#> Variables: goal12 and ef_legal, Correlation: -0.837627#> Variables: goal16 and ef_legal, Correlation: 0.839944#> Variables: pf_law and ef_legal, Correlation: 0.852443# List of high-correlation pairscorrelation_pairs <-list(c("overallscore", "goal1"), c("overallscore", "goal3"), c("goal1", "goal3"),c("overallscore", "goal4"), c("goal1", "goal4"), c("goal3", "goal4"),c("overallscore", "goal6"), c("goal3", "goal6"),c("overallscore", "goal7"), c("goal1", "goal7"), c("goal3", "goal7"), c("goal4", "goal7"), c("goal6", "goal7"),c("overallscore", "goal9"), c("goal3", "goal9"),c("overallscore", "goal11"), c("goal3", "goal11"), c("goal4", "goal11"), c("goal6", "goal11"), c("goal7", "goal11"),c("goal9", "goal12"), c("goal12", "goal13"),c("overallscore", "goal16"), c("goal12", "goal16"),c("goal9", "GDPpercapita"), c("goal12", "GDPpercapita"),c("overallscore", "internet_usage"), c("goal9", "internet_usage"),c("goal12", "pf_law"), c("goal16", "pf_law"),c("pf_religion", "pf_assembly"), c("pf_assembly", "pf_expression"),c("goal9", "ef_legal"), c("goal12", "ef_legal"), c("goal16", "ef_legal"), c("pf_law", "ef_legal"))# Flatten the list and count the frequency of each variableflattened_list <-unlist(correlation_pairs)frequency_count <-table(flattened_list)variables_to_remove <-c()for (pair in correlation_pairs) {# Select the variable that appears more frequently for removalif (frequency_count[pair[1]] > frequency_count[pair[2]]) { variables_to_remove <-c(variables_to_remove, pair[1]) } elseif (frequency_count[pair[1]] < frequency_count[pair[2]]) { variables_to_remove <-c(variables_to_remove, pair[2]) } else {# If both appear equally, arbitrarily choose one to remove variables_to_remove <-c(variables_to_remove, pair[1]) }}variables_to_remove <-unique(variables_to_remove)variables_to_remove <-sort(variables_to_remove)print(variables_to_remove) #> [1] "ef_legal" "goal11" "goal12" "goal16" #> [5] "goal3" "goal4" "goal7" "goal9" #> [9] "overallscore" "pf_assembly"
::: Therefore, we will not take into account the variables “ef_legal” “goal11” “goal12” “goal16” “goal3” “goal4” “goal7” “goal9” “overallscore” “pf_assembly” in our regressions, for multicollinearity purpose.
Now, let’s compute the regressions without these variables. ::: {.cell layout-align=“center” hash=‘report_cache/html/unnamed-chunk-301_a97da02a661143bc4a5337119c40e152’}
::: The problem is that even by getting rid of the previous variables, there still might be multicollinearity. Therefore, we need to analyse the vif for each regression and adapt the model in consequence. ::: {.cell layout-align=“center” hash=‘report_cache/html/unnamed-chunk-302_981b3fe3887d799216f41c444022018a’}
##### geom point ######print values with correlation > 0.8 and make plots# Filtering values where the absolute value is greater than 0.8highcorrelations <- melted_corr_matrix_GVar %>%filter(value >0.8)ggplot(data_question1, aes(internet_usage, overallscore)) +geom_point()+geom_smooth(se =FALSE) +labs(title ="Scarplot overallscore and internet usage")ggplot(data_question1, aes(GDPpercapita, goal9)) +geom_point()+geom_smooth(se =FALSE) +labs(title ="Scarplot overallscore and internet usage")ggplot(data_question1, aes(internet_usage,goal9)) +geom_point()+geom_smooth(se =FALSE) +labs(title ="Scarplot overallscore and internet usage")ggplot(data_question1, aes(ef_legal,goal9)) +geom_point()+geom_smooth(se =FALSE) +labs(title ="Scarplot overallscore and internet usage")ggplot(data_question1, aes(pf_law, goal16)) +geom_point()+geom_smooth(se =FALSE) +labs(title ="Scarplot overallscore and internet usage")ggplot(data_question1, aes(ef_legal, goal16)) +geom_point()+geom_smooth(se =FALSE) +labs(title ="Scarplot overallscore and internet usage")
=======
Code
##### geom point ######print values with correlation > 0.8 and make plots# Filtering values where the absolute value is greater than 0.8highcorrelations <- melted_corr_matrix_GVar %>%filter(value >0.8)ggplot(data_question1, aes(internet_usage, overallscore)) +geom_point()+geom_smooth(se =FALSE) +labs(title ="Scarplot overallscore and internet usage")ggplot(data_question1, aes(GDPpercapita, goal9)) +geom_point()+geom_smooth(se =FALSE) +labs(title ="Scarplot overallscore and internet usage")ggplot(data_question1, aes(internet_usage,goal9)) +geom_point()+geom_smooth(se =FALSE) +labs(title ="Scarplot overallscore and internet usage")ggplot(data_question1, aes(ef_legal,goal9)) +geom_point()+geom_smooth(se =FALSE) +labs(title ="Scarplot overallscore and internet usage")ggplot(data_question1, aes(pf_law, goal16)) +geom_point()+geom_smooth(se =FALSE) +labs(title ="Scarplot overallscore and internet usage")ggplot(data_question1, aes(ef_legal, goal16)) +geom_point()+geom_smooth(se =FALSE) +labs(title ="Scarplot overallscore and internet usage")
>>>>>>> Stashed changes
Let’s explore how the different SDG are correlated together by creating a heatmap of the correlation between our variables. We added a script to check whether the correlations are significantly different from 0. First, let’s select the SDGs scores.
We then reshape our data to be able to use the ggplot2 package to create our heatmap.
Code
melted_cor_matrix <-melt(cor_matrix)melted_p_matrix <-melt(matrix(as.vector(p_matrix), nrow =ncol(sdg_scores)))plot_data <-# Combine the datasetscbind(melted_cor_matrix, p_value = melted_p_matrix$value)ggplot(plot_data, aes(Var1, Var2, fill = value)) +geom_tile() +geom_text(aes(label =sprintf("%.2f", value), color = p_value <0.05),vjust =1) +scale_fill_gradient2(low ="blue", high ="red", mid ="white", midpoint =0, limit =c(-1,1), space ="Lab", name="Pearson\nCorrelation") +scale_color_manual(values =c("black", "yellow")) +# black when significant, yellow if nottheme_minimal() +theme(axis.text.x =element_text(angle =45, hjust =1),axis.text.y =element_text(angle =45, hjust =1),legend.position ="none") +labs(x ='SDG Goals', y ='SDG Goals',title ='Correlation Matrix with Significance Indicator')
>>>>>>> Stashed changes
Note that as said previously, we assessed the correlations to ascertain if they substantially deviated from zero, setting the significance level at an alpha of 5%. To aid in visualization, we marked any correlations that did not meet this level of significance with a yellow highlight in our graphical representation. The absence of yellow markings on our plot suggests that all Sustainable Development Goal (SDG) scores demonstrate a statistically significant correlation.
We can have a look at the shape of the corelation between the SDGs with the plot function.
